WORKING PAPER
2001
Department of Economics
Tufts University
Medford, MA 02155
(617) 627 – 3560
http://ase.tufts.edu/econ
RESEARCH SEMINAR IN INTERNATIONAL ECONOMICS
School of Public Policy
The University of Michigan
Ann Arbor, Michigan 48109-1220
Discussion Paper No. 474
CHILD LABOR:
THEORY, EVIDENCE AND POLICY
Drusilla K. Brown
Tufts University
Alan V. Deardorff and Robert M. Stern
University of Michigan
August 17, 2001
Recent RSIE Discussion Papers are available on the World Wide Web at:
http://www.spp.umich.edu/rsie/workingpapers/wp.html
ii
CHILD LABOR:
THEORY, EVIDENCE AND POLICY
By
Drusilla K. Brown
Tufts University
Alan V. Deardorff
University of Michigan
Robert M. Stern
University of Michigan
August 17, 2001
Send correspondence to: Drusilla K. Brown
Department of Economics
Tufts University
Medford, MA 02155
e-mail: [email protected]
phone: 617-627-3096
fax: 617-627-3917
iii
Paper: EGDI-WP.doc
ABSTRACT
CHILD LABOR:
THEORY, EVIDENCE AND POLICY
Drusilla K. Brown
Tufts University
Alan V. Deardorff
University of Michigan
Robert M. Stern
University of Michigan
There is a growing theoretical and empirical literature concerning the causes and
consequences of child labor. The objective of this paper is to evaluate the policy initiatives
targeted on child labor in light of the newly emerging theoretical argumentation and empirical
evidence. We focus in particular on programs to address child-labor practices, and we attempt to
evaluate these programs, given the empirical evidence concerning the primary determinants of
when and why children work. Throughout, we find it instructive to evaluate the policies that have
been adopted with the intent of reducing overall child labor in terms of the impact they are likely
have on the welfare of children.
Keywords:
Child labor
Labor standards
JEL Subject Code:
J2 Employment Determination
F13 Commercial Policies
Correspondence:
Drusilla K. Brown
Department of Economics
Tufts University
Medford, MA 02155
Tel. 617-627-3096
Fax. 617-627-3917
E-mail: [email protected]
1
August 17, 2001
CHILD LABOR:
THEORY, EVIDENCE AND POLICY
Drusilla K. Brown
Tufts University
Alan V. Deardorff and Robert M. Stern
University of Michigan
I.
Introduction
Over the past decade, child-labor practices in developing countries and their implications
for international trade have become a focus for attention in the international arena. Pursuant to
these concerns, some developing countries have adopted innovative programs designed to
discourage the worst child-labor practices and to provide families and communities with
incentives to reduce child labor and increase educational attainment. Additionally, some
industrialized countries have considered a range of policy options, both punitive and nonpunitive, intended to reduce imports of goods produced by children and to provide incentives and
financial support to reduce child labor in traded and non-traded sectors. Given the increased
attention to child labor and the threat of trade sanctions by industrialized countries for weak child
labor protections, it is instructive to evaluate the policies that have been adopted with the intent of
reducing overall child labor in terms of the impact they are likely have on the welfare of children.
There is a growing theoretical and empirical literature concerning the causes and
consequences of child labor. The overriding objective of this paper is to evaluate the policy
initiatives targeted on child labor in light of the newly emerging theoretical argumentation and
empirical evidence. We will focus in particular on programs to address child-labor practices, and
we will attempt to evaluate these programs, given the empirical evidence concerning the primary
determinants of when and why children work.
2
Particular attention will be given to the causes of child labor above and beyond poverty.
It is widely accepted that child labor declines as per capita income rises. However, the process of
economic development is a slow one, and many developing countries have lost ground over the
last decade both in terms of their standard of living and progress made in reducing child labor.
Therefore, we would like to focus particularly on the other causes of child labor distinct from
poverty and the policy remedies that theory and evidence suggest.
In Section II, we begin with some of the theoretical arguments concerning family
decision-making and the determinants of child labor. We then turn in Section III to the empirical
evidence concerning the determinants of child labor and their implications for the types of
policies that are likely to influence household decision-making in a manner that reduces the
incidence of child labor and increases educational attainment. Section IV is devoted to a review
of the traditional methods for reducing child labor. Section V provides an overview and
discussion of the likely effectiveness of recent initiatives targeting child labor by developing
country governments and initiatives underway by international agencies such as the World Bank
and International Labour Organization. In Section VI, we examine some of the motivations for
including child labor on the international trade agenda and the likely implications of doing so.
Conclusions follow.
II. Theories of Child Labor: Models of Household Decisions
The purpose of this section is to touch briefly on theories of household decision-making
with regard to the employment of children. Greater emphasis will be placed on the more recent
literature that addresses the role of market failure, particularly in the capital market, and its
relationship to poverty. The ultimate objective of the review is to identify the household
characteristics that ought to emerge in empirical analysis as statistically significant determinants
of child labor.
3
Neoclassical models of household decision-making are commonly employed in the
analysis of child labor and are typically derivative of Becker (1964). Models of household
bargaining fall into two broad categories: those in which children have no bargaining power and
those in which children have some intrinsic value in the family. In models in which children have
no bargaining power in the household, parents make decisions that serve their own interests,
without regard for the impact on the child. This class of models lends analytical support for
public policies that constrain the choices that parents are allowed to make for their children, eg.,
compulsory schooling, minimum age of work, a ban on bonded child labor, etc.
Children as Household Assets
In this context, children are viewed strictly in terms of their value as assets. Parents first
must choose the number of children they will have. They then weigh whether to invest in the
quality of the child or to extract a current stream of services. Becker and Lewis (1973) argue that
in the quality-quantity tradeoff, parents who choose a large number of children are less likely to
invest in quality schooling. That is, the number of children and investment in the human capital
of children are substitutes. Or, parents may choose to have a large number of children in order to
diversify risk, formally educating some and putting the others to work
Initial empirical analysis was quite supportive of both the quality-quantity trade-off and
the diversification hypothesis. Rosenzweig and Wolpin (1980) find that an exogenous increase in
fertility lowers child quality, and Hanushek (1992) finds a trade-off between family size and
educational attainment in the United States. Indeed, there is considerable evidence that, on
average, children in larger families in both developed and developing countries receive less
schooling, perform more poorly on intelligence tests, and are less well nourished (Patrinos and
Psacharopoulos, 1997). Closely-spaced children receive the least investment (Powell and
Steelman, 1993).
4
However, Montgomery, Kouame and Oliver (1995) find contradictory evidence for
Ghana and Cote d’Ivoire. Further, Patrinos and Psacharopoulos (1995) do not find that the
number of siblings of Paraguayan children affect the level of enrollment. Nor is there a strong
sibling effect in Brazil (Psacharopoulos and Arrigada, 1989; Levison, 1991). Chernichovsky
(1985), studying schooling choice in rural Botswana, actually finds that family size raises
educational attainment. Levison (1991) suggests that the positive correlation between family size
and schooling may occur because there are decreasing returns in household production. With a
large number of children available to engage in household work, the opportunity cost of
education for any one child may be quite low.
Not only are child assets viewed in relation to one another, the labor of children may also
be seen as a complementary input to other household capital. For example, the investment in
physical capital to launch a family enterprise may be optimizing only if it can be combined with
the labor of the household’s children.
In fact, exploiting a household’s assets may frequently require the inputs of child labor.
One of the most well documented cases of complementarity concerns the work and school habits
of girls whose mothers have marketable skills. Tapping the mother’s human capital in the formal
labor market may require her daughters to replace her contribution to home work. Thus, human
capital embodied in a mother is complementary with more home work and less education for her
daughters.
The Poverty Hypothesis
Although parental selfishness may play a role, it is a very commonly held view that child
labor is fundamentally a by-product of poverty, strongly suggesting that policy should focus on
economic development and increasing income (Nardinelli, 1990). Krueger (1996) notes a steep
cross-country negative correlation between GDP per capita and the employment rate of 10-14
year olds in 1995. An important implication of the poverty hypothesis is that policies that focus
5
on compelling parents to deviate from their optimizing choices may, in fact, make children worse
off.
Although the poverty/child-labor link may seem obvious, Baland and Robinson (2000)
formalize this idea, thus helping to isolate the precise nature of the mechanism. They take as a
point of departure that all families make child-labor decisions to maximize the present discounted
value of the household’s income. In making child-employment decisions, parents weigh the
present discounted value of the future income of an educated child against the foregone income
while the child is in school. Child labor is only chosen if the return to education is not high
enough to compensate families for the lost income of their children.
The obvious question then becomes “what is it about being poor that lowers the present
discounted value of an education relative to current work?” In a world with perfectly functioning
capital markets there are two possibilities: (1) poor people are impatient; that is, they discount the
future more heavily than other families; and (2) the return to education for a poor child is lower
than for children generally. A low return to education for poor children will occur if schools are
far away, inadequately staffed, lack educational supplies and materials, etc. The return to
education may also simply be unappreciated if the parents themselves are not educated.
A third possibility emerges when the parent’s initial endowment is low relative to their
child’s future income (whether or not they are educated). In this case, parents would like to
engage in consumption smoothing. That is, they would like to borrow against the household’s
future wealth to increase current consumption while lowering future consumption, thereby
evening out the consumption profile of the family over time. Parents are particularly highly
motivated to engage in consumption smoothing when the household’s survival is threatened by a
period of unemployment, drought, etc.
In such a case, parents could offer the child a deal whereby the parent borrows on behalf
of the current household, expecting the child to provide the funds to repay the loan out of the
income they will earn as an educated adult. The problem is that a child cannot pre-commit to
6
compensate the parents from future income. Thus, the only option parents have for increasing
current household consumption at the expense of the future is to put the child to work.
Baland and Robinson note that this type of bargaining failure occurs when the optimal
bequest is negative. That is, over time, it is optimal (from some point of view) to transfer
resources back in time from the future of the household to the present, rather than from the
present to the future. Implicit in the Baland-Robinson analysis is the fact that child labor is a
device for transferring income from the future into the present. A child who works today at the
expense of acquiring an education will contribute to family income today at the expense of future
productivity.
Evidence of intra-household bargaining problems of the sort raised by Baland and
Robinson (2000) are found by Parsons and Goldin (1989) and Andvig (1997). Both studies find
that children leave the household after receiving an education, making it difficult for parents to
internalize the benefits of investing in their children. Further, and perhaps more to the point,
Parsons and Goldin (1989) find from their analysis of the U.S. 1900 Census, that working
children received little of their earnings in the form of bequests. Child wages only raised current
household consumption. One way to interpret this result is that working children were
transferring income back to their parents. That is, the optimal bequest was negative, which is
precisely the situation in which Baland and Robinson expect to observe working children.
It is important to note that while Baland and Robinson find an analytical role for poverty
as a source of child labor, their analysis does not suggest that we should rely exclusively on
economic development as a strategy for eradicating child labor. Rather, as we will see below,
government policy can play a significant role in solving the intra-household bargaining problem.
Market Failure and Multiple Equilibria
A second wave of models assumes that parents are altruistic and focuses on the
interaction between market characteristics and child labor that point to certain market
7
manipulations as a remedy. For example, Basu (1999) examines the case in which rigidities in
the market for adult labor drive child labor. He considers a market in which the adult wage is
downward rigid, giving rise to adult unemployment. For those families with an unemployed
adult, children must work. A policy aimed at restoring wage flexibility and improving labormarket function might lower child labor, as would a subsidy to the household during a period of
unemployment.
Grootaert and Kanbur (1995) focus attention on the external benefit of an educated child
on the general population. In this case, an education subsidy will help internalize the externality
and may have the added benefit of reducing child labor.
Basu and Van (1998) analyze the case in which benevolent parents withdraw their
children from the labor market once the adult wage reaches a critical level. Such a labor market
may have two equilibria: one in which both children and adults work, giving rise to a large labor
supply and low wages, and one in which only parents work, giving rise to a low labor supply and
a high adult wage. A ban on child labor may have the effect of helping the high-wage, no-childlabor equilibrium emerge, thus redistributing income towards the supplying families and away
from owners of other factors.
Hirshman1 also suggests the possibility of multiple equilibria emerging when parents who
put their children to work suffer a social stigma for doing so. The stigma is greater, the smaller
the number of other children in the community who are working. Thus, as with Basu and Van,
there may be two equilibiria: one with many children working and a low social stigma attached to
work and one with few children working and a high social stigma. Hirshman’s analysis also
suggests that a policy banning child labor might bring about the low child-labor/high socialstigma equilibrium.
The most recent developments focus on capital market failure. Baland and Robinson
(2000) extend their approach discussed above, introducing the possibility that households might
8
be liquidity constrained. Thus, they emphasize the importance of capital-market failure as a
contributing factor to inefficient child labor.
Capital-market failure can emerge in several different guises. Consider first the case in
which the present discounted value of an education is greater than the current value of a child’s
labor. In this case, it is clearly optimal for a family to borrow against the child’s future income to
finance the child’s education. Or, to be more specific, it is in the interest of the child to make any
requisite contribution to household income by borrowing against future income, thus freeing the
child to attend school rather than work. Baland and Robinson note that the inability of the child
to access the capital markets, or the inability of the child to pre-commit to repay education loans
obtained by the parents on the child’s behalf, may give rise to inefficiently low educational
attainment.
To the extent that such intra-family bargaining failure is contributing to inefficient
educational attainment, it is possible for government policy to correct the failure with properly
configured educational loans to poor families. A government loan that is tied to the child’s
educational performance and becomes the liability of the child, rather than the parent, allows the
child to access the capital markets to meet required contributions to the family. Such a loan is
Pareto improving provided there is some reason to believe that the child would have voluntarily
undertaken the loan if he/she had the cognitive ability to analyze the choices like an adult.
However, if parents treat their child’s future as a contributing factor to their own sense of
well-being, they may be willing to borrow against their own assets or future income in order to
finance their child’s education. In this case, a lack of collateral will prevent parents from
accessing the capital markets, thus again giving rise to an inefficiently low level of education.
The dynamic implications of capital market failure have been studied by Ranjan (2001),
with similar conclusions reached by Basu (1999). Ranjan considers, in particular, very poor
families who would choose to educate their children if they had access to a capital market, but fail
1
See Basu (1999)
9
to do so due to capital-market failure. Such families produce poor, uneducated children who
repeat the cycle for the next generation. In this model, a concerted effort to educate one
generation of such children will pull the family away from the income level at which they depend
on the labor of their children for survival. Thus, subsequent generations of the family will be able
to educate their children, permanently enjoying a higher standard of living and educational
attainment.
The central policy lesson of the Ranjan and Basu models is that government intervention
is required for only one generation of children. For, once an educated child’s future income is
raised above a threshold level, the newly created parent will be able to choose education rather
than child labor for the next generation.
III. Empirical Evidence on the Determinants of Child Labor
The purpose of this section is to review the empirical evidence on the determinants of
child labor. The theoretical models considered above suggest several potential motivations for
putting children to work. As noted in each case, knowing the cause of child labor is fundamental
to making effective policy.
Evidence to support a view of selfish parents is provided by Burra (1995) and Parsons
and Goldin (1989). Moehling (1995) also finds that the bargaining position of the children in the
household varies with the child’s contribution to income. Gupta’s (1997) analysis of working
children in West Bengal, India, suggests that children have very little bargaining power in the
household. Thus, there is some evidence of at least a little play in the resources that a family can
devote to the welfare of their children. At the margin, properly configured policy may push some
families to increase the investment they make in their children.
10
Household Decision-Making and Child Labor
In order to disentangle the conflicting determinants of child labor and to assess the
relative importance of each of the factors influencing child-labor decisions, it is necessary to
empirically estimate household decision-making in the context of a formal analytical model.
In the most careful of such studies, households are assumed to use a sequential process
for making child-labor decisions. In the first stage, parents decide whether a child should work.
In the second stage parents decide whether the child will work part-time or full-time. In the third
stage, the type of work is chosen. Sequential probit analysis is undertaken on household survey
data to identify the family characteristics that are determining the probability that a child works,
the probability of schooling, and the type of work.
Typically, such analysis will begin with the specific characteristics of the child such as
age and gender. Parental characteristics such as educational attainment, age, and employment
status are also included.
Household characteristics include, first and foremost, some measure of household
income. Due to problems with endogeneity of income, most analysts include measures of
household capital, welfare, poverty status, total expenditures or expenditure on food, in lieu of
household income.
Household assets are also important in the absence of access to formal capital markets.
Households that want to borrow against the future may be able to tap internal assets. The
presence of the father in the household, the presence of an older sibling in the household
(particularly a brother), the capacity of the mother to engage in market work, or property
associated with a family enterprise can all be thought of as assets that can be drawn upon even if
the family has no access to formal capital markets. For this reason, the presence of such
household assets might be expected to lower child labor. Consequently, gender, birth order, the
presence of older siblings, the mother’s work opportunities, and the presence of a family
11
enterprise are also important determinants of whether a child works, the type of work undertaken,
the number of hours worked, and whether part-time schooling is an option.
The availability of schools in terms of quality, proximity and cost will also affect child
labor and schooling choices. Household expenditures on schooling are typically available from
survey evidence. However, measuring school quality is extremely difficult. At best, some
studies have evidence on the integrity of the school structure, whether or not the school is open
most days of the week, and other services available to the general community such as running
water or electricity.
Nearly every study includes some other characteristics such as region of the country,
urban vs. rural, and other cultural characteristics including religion.
Determinants of Work and School
Consider first the decision as to whether children should work at all. Empirical results
for Colombia, Bolivia, and Peru are reported in Table 1. Each study uses a slightly different set
of explanatory variables. Those significant at the 10% level are listed in the corresponding
columns. Regression results reported in Table 2 are probability derivatives evaluated at the mean
of the explanatory variables.
Cartwright (1999) analyzes 1993 survey data for rural and urban children in Colombia.
As we will see below, gender usually plays a significant role in work and school decisions, but
the role of gender varies across continents. In the case of Colombia, rural boys are 9% more
likely to work than are girls. The age of the child is also significant in virtually every country
studied. In Colombia, the probability of work increases by 8 percentage points for each year a
child ages. As the child ages and becomes more productive, the opportunity cost of education
rises, making work more attractive
12
Poverty also plays a substantial role in driving children to work. For a one percent
increase in household expenditure in Colombia, the probability of work declines by 0.11
percentage points for a rural child and 0.19 percentage points for an urban child.
Also, as expected, household assets are an important determinant of whether children
work, although the direction depends on the nature of the asset. The presence of a household
enterprise for rural families makes it easier to draw children into work. Children in rural
Colombia whose families operate a household enterprise are nearly 8 percent more likely to work
than other rural children. The fact that the mother is working may also make child labor in the
household necessary. Rural children are 7 percent more likely to work and urban children are 4
percent more likely to work if their mothers also work. These are household assets that are most
effectively tapped when children in the household work. Thus, for these two household assets,
the asset and child labor are complementary.
The role of siblings is particularly interesting. There is little evidence that siblings
increase child labor supply. The presence of younger siblings does not affect the probability of
working, so older children do not appear to be engaged in childcare. However, the presence of
other siblings in the same age range plays a significant role in lowering the probability of work,
and the presence of older children in the household also lowers the probability of any one child
working.
These results certainly provide very little evidence for a diversification in the investment
in children. Siblings within an age-range, 10-15 years old, increase the probability that each other
are in school. This is far more consistent with the notion of decreasing returns to household
production than diversification.
One possible way to preserve the diversification hypothesis is to assume that parents with
a single child diversify by having the child work part-time and in school part-time. Once
additional children are born, diversification can be accomplished by putting some children to
work exclusively and some in school exclusively. If this is the case, the presence of additional
13
siblings should have the twin effects of increasing the probability of full-time schooling and
increasing the probability of full-time work. That is, the larger the family, the less likely we are
to observe a work-school combination.
However, as we will see below, the presence of other siblings in the household also
typically lowers the probability of full-time work. The negative impact of other siblings in the
household on full-time work is particularly notable for urban Colombia. The only exception is
that very small brothers in rural Colombian households increase the probability that their older
siblings will work full-time. Cartwright was somewhat skeptical of this result because the
presence of very young sisters lowers the probability of full-time work for older rural Colombian
siblings. She was unable to account for the distinction.
Nor does there appear to be evidence of a quantity-quality tradeoff. The only possible
way to interpret the results in favor of the trade-off hypothesis is that parents are putting their
first-born to work and then investing in formal education for the younger children. However, this
configuration defies conventional wisdom. Hanushek (1992) and others argue that appearing
high in the birth order has significant advantages. Parents typically invest in the first child since
first-borns have a higher probability of being in a small family than subsequent children.
It seems far more compelling to interpret the presence of older siblings in the household
as evidence of household capital on which the parents can draw in lieu of tapping formal capital
markets, making it possible to keep younger children in school full-time. This is an important
conclusion because it lends support for the notion that improved access to capital markets for
families with limited household assets might reduce the incidence of child labor.
Furthermore, the evidence also suggests that increasing the size of each cohort of children
in the family, thereby increasing sibling density, lowers the probability of child work. We would
expect the opposite if parents with a large number of closely spaced children were planning to put
them to work rather than in school. Thus, we have far more evidence of decreasing returns to
14
household production rather than a quality-quantity tradeoff. Children with a large number of
similarly aged siblings share household chores and then also attend school.
We will also find that parental education plays a persistent, powerful and negative role in
the family’s decision to put a child to work. The more years of school both mothers and fathers
have, the more likely they are to devote their children’s time exclusively to school, even
controlling for household income. This effect is more ubiquitous than any other in determining
child labor. In the case of Colombia, as noted in Table 1, the parental education effect is
particularly pronounced. Each year of each parents’ education lowers the probability that their
child will work full time by 2 percentage points in rural Colombia.
Note that the parental education effect exists above and beyond the contribution that the
educated parent’s human capital makes to family income. Thus, when parents become educated,
this appears to impart some informational externality that affects the decisions that parents make
for their children.
Finally, the cost of schooling in Colombia has a negative effect on child labor. That is,
the more expensive school is, the less likely it is that the child will work. Cartwright suggests
that the cost of schooling in this case is a proxy for school quality.
Results for the second stage of the estimation are reported Table 2. The second stage
includes only those children who are working and attempts to determine which of these children
work full-time.
For children who work, the decision to work full time depends on many of the same
variables as the decision whether to work at all, although there are some important exceptions and
nuances. For example, even though boys are more likely to work, girls are more likely to work
full-time, both in market work and household production. Further, there is some evidence that
when children are taken out of school to work full-time, they are caring for very young male
siblings.
15
It is also important to note the subtle role that a family enterprise and mother’s work have
on children in Colombia. Both of these household assets are complementary with some child
labor. However, they also appear to make it possible to combine work and school. So, children
in families with a household enterprise and/or a mother who works are more likely to be working
than other children, but they are not put to work full-time. A household enterprise gives the
family some flexibility in when and how hard their children work and also allows parents to
supervise their children themselves. In such cases, child labor may not seem so egregious and
may more readily be combined with formal schooling.
Cartwright and Patrinos (1999) find somewhat similar results for Bolivia as shown also in
Tables 1 and 2. Poverty plays a central role in driving child labor. The effect of poverty is
mitigated only by a mother’s educational attainment and the presence of other older siblings in
the household. However the cost of schooling is also key. The more expensive schooling the
more likely children are to work, as one would expect.
But, for Peru, as analyzed by Patrinos and Psacharopolous (1997), we have a dramatically
different story. As can be seen from the last column of Table 1, boys are more likely to work
than girls, and the father’s education has a small negative impact on the decision to work. There
is also some evidence that older children are caring for younger children, but this effect does not
appear to be driven by the mother’s employment.
What is noteworthy in Peru is that none of the potential measures of household assets or
income appear to play a significant role. This is quite a surprising result in light of the strongly
held popular notion that child labor is most importantly a consequence of poverty. (Similar
results for Peru are found by Ray (2000)).
In order to tease out the more subtle aspects of Peruvian child labor, Patrinos and
Psacharopolous also try to determine why a child fails to advance in school. They use as a
dependent variable a measure of the age-grade distortion, which indicates the extent to which a
16
child is failing to advance through school with his/her cohort. Results are reported in the last
column of Table 2.
Neither family income nor whether the child is employed played a role in success in
school. Neither of these variables is statistically significant in explaining the age-grade
distortion. The only income or asset variable that does appear to be important is the number of
rooms in the home. Children who fall behind in school have a large number of siblings who are
also in school. Thus, child labor and educational attainment do not appear to be connected in
Peru, with only one exception. Older Peruvian children are, in some cases, taken out of school to
care for younger children. Otherwise, the type of work engaged in by Peruvian children does not
appear to be a competitor with schooling for a child’s time or intellectual energy. Consequently,
it is not surprising that income and household asset variables play little role in driving the
decision to work.
Patrinos and Psacharopoulos hypothesize the reason that Peruvian children can combine
work and school without ill effect is that work makes it possible to afford to attend school.
However, it is equally possible that these results point to the poor quality of school. Perhaps the
value of schooling is so low that parents do not see school attendance in financial terms.
The role of poor school quality should not be understated. Peru provides a particularly
striking example. The deficiencies in the facilities, supplies, teacher salaries, and training
seriously undermine the value of the time that children spend in school. A third of all schools
have only one teacher, a problem most common in rural areas. In rural Amazon, it is typically the
case that there is only one teacher to handle 50-60 students in 4-6 different grades. The poorest
schools may not have even the most fundamental educational supplies such as books. In rural
areas, 83 percent of schools have no running water, electricity or indoor plumbing. Even in
metropolitan Lima, only 60 percent of schools have electricity (Ministerio de Educacion, 1993).
17
Given the deficiencies in the public education system, some children work for the explicit
purpose of earning the tuition for private education. For example, in Ecuador, one in ten working
children studies in a private school.
It is also important to note that the weak impact of child labor on academic performance
in Peru is not evident elsewhere. For example, Psacharopoulos (1997) finds that a working child
in Bolivia is 10 percent more likely to fail a grade than an unemployed child. Similar results are
obtained for Venezuela.
Turning to evidence from Africa, we find several differences, one of which is that the
cultural attitudes toward gender, work and school vary across continents. Canagarajah and
Coulombe (1998), analyzing the work and school choices in Ghana, find that girls are more likely
to work than boys and less likely to attend school, as can be seen from the first column of Table
3. The differential rate is directly attributable to homework performed by girls. However, as
with Peru, household income does not play a strongly positive role in whether children work. In
fact, the correlation between income and child labor follows an inverted-U shape. Thus, child
labor falls with income only in the upper range of the income distribution. This relationship is
confirmed by Bhalotra and Heady (1998) and Levison (1991).
We do see evidence that family assets influence school and work decisions. The
presence of a non-farm family enterprise and older women in the household lower the probability
of work for younger children. However, children are more likely to work if the parents are selfemployed farmers. Years of parental education and the presence of older siblings also lower the
probability of work and raise the probability of some schooling.
Interestingly, school fees increase both the probability of work and the probability of
school attendance. This strongly suggests both that children need to work to pay school fees, but
that more expensive schools are higher quality.
Given the indication that household assets, as distinct from income, are playing a role in
child labor decisions, it would be useful to have some more direct evidence from financial assets
18
rather than simply trying to draw inferences from the presence of human capital. Jensen and
Nielsen (1997) include several asset measures in their study of schooling decisions in Zambia.
Results are presented in the last column of Table 3. As in Ghana, girls are less likely to be in
school than boys. This is particularly the case for girls who are married.
We also see some other familiar results. For example, educated fathers are more likely to
school their children. Families with a wage-earner are also more likely to school their children.
However, children in households with more than one family member working other than the child
are considerably less likely to be in school.
Income, again, has a strong positive effect on schooling, as do financial assets and
household capital. However, if the family holds its assets in the form of land, children are less
likely to be in school. Land and child labor appear to be complementary inputs in Zambia. Note,
though, that children in the rural sector and whose family members work in the agricultural sector
are more likely to be in school. It may be that such families have more flexibility in combining
agricultural work and schooling. However, Jensen and Nielsen also suggest that rural schools
may be of higher quality than urban schools, thus making schooling a more attractive choice for
agricultural families.
Much of the same character of child labor in Zambia is evident for Cote d’Ivoire, as well.
Grootaert (1999) uses sequential probit and bivariate techniques to analyze the sequence of child
labor decisions for children aged 7 to 17. Results for urban children are reported in the middle
columns of Table 3. Grootaert first isolates the household characteristics for families that send
their children to school only. Next, of those children who do some work, children are sorted
between those who combine work and school and those who work only. Finally, of those who
only work, children are divided between market/family enterprise workers and home-workers.
Several of the standard results are in evidence in Grootaert’s estimates. Girls are 30
percent less likely to be in school full-time and are also less likely to combine work and school
and less likely to work in the market than boys.
19
Educated parents are more likely to send their children to school full-time or to combine
work and school than to put children to work only. Each year of a father’s education lowers the
probability of dropping out of school by 1.8 percentage points, and each year of a mother’s
education lowers the probability of dropping out by 3.5 percentage points. In fact, parental
education is more important than any other variable in deterring full-time employment by
children. For rural children the impact is even more pronounced. (See original manuscript for
rural results). Each year of a father’s education lowers the probability of dropping out by 7
percentage points and each year of a mother’s education lowers the probability of dropping out by
3 percentage points. However, in both cases, the effect is weaker for girls than for boys.
Poor families are less likely to send their children to school or combine school with work.
Urban families in the lowest income quintile are nearly 9 percent less likely to send their children
to school full-time and over 12 percent less likely to have children combine school with work
rather than drop out completely.
Grootaert argues that his results suggest a strong role for income in determining child
labor. However, he does not enter income directly into the equation due to problems with
endogeneity, preferring instead to rely on measures of household capital. Coulombe (1998),
evaluating a similar data set, finds that income, corrected for endogeneity, plays only a small role.
Each one percent increase in household income lowers the probability of child labor by 0.3
percent.
The impact of household assets on child work, once again, depends on the nature of those
assets. In Cote d’Ivoire, the presence of a family enterprise lowers the probability of school only
and the probability of a work-school combination. One possible explanation is that a family that
decides to invest in a family enterprise is doing so with the expectation that the children will work
in the family enterprise and thus will not need an education.
Grootaert suspects that the household enterprise is a proxy for poverty, as most household
enterprises in Cote d’Ivoire are run by very poor families. However, some corroborating
20
evidence of the complementarity between child labor and household assets is offered by
Coulombe (1998). In rural Cote d’Ivoire, the probability of working increases and the probability
of school attendance decreases with the number of acres of land owned by the household.
The family enterprise also interacts in an unusual way with the work of mothers and
daughters. In Cote d’Ivoire, it appears to be the case that mothers who work in a family
enterprise draw their daughters in, working side-by-side both in the enterprise and in household
production. Otherwise, unlike in some other studies, a working mother increases the probability
of full-time schooling.
In Cote d’Ivoire we see another by now familiar result. The presence of other work-age
children in the household increases the chances of schooling and lowers the probability of work at
all stages of decision-making. The distance to school also plays a role for rural children. The
absence of a school in the local community increases the probability of dropping out for rural
children by 18 percentage points.
Several studies above have pointed to the importance of school quality as a determinant
of child labor. However, few studies incorporate any measure of school quality. Dreze and
Kingdon (2000), in their study of child labor in India, do not have measures of school quality, but
they do have some indication of desirable school characteristics that might be correlated with
school quality, such as whether the building’s roof is waterproof. They also include evidence as
to whether the community has made a commitment to building infrastructure, such as whether
homes have electricity, piped water and phone service. The impacts of these variables on
probability of enrollment for boys and girls are reported in the first two columns of Table 4.
(Similar results are found for Pakistan by Ray (1998)).
Although factors such as school quality and village characteristics do not impact
schooling decisions for boys, they are quite important for girls. In addition, the availability of a
school-lunch program, which increases the short-term return to school attendance, and a positive
attitude toward school attendance both increase the probability of school enrollment for girls.
21
Glewwe and Jacoby (1994), in their analysis of Zambia, also find an important role for
school quality. For example, they find that repairing classrooms in schools that are unusable
when it rains is more valuable than providing additional instructional material. Hanushek and
Lavy (1994) find that Egyptian students attending higher quality schools tend to stay longer in
school and complete more grades. Finally, San Martin (1996) finds that labor-force participation
(LFP) rates for children aged 10-14 rise with the primary-school, student-teacher ratio across
countries.
The Demand for Child Labor
Up to this point we have focused on the supply side of child labor. On the demand side,
it is sometimes argued that the demand for children derives from their special characteristics that
make them indispensable for production; the so-called “nimble fingers” argument. Some
employers claim that only children with small fingers have the ability to make fine hand-knotted
carpets, pick delicate jasmine flowers, or scramble through narrow tunnels. However, like the
argument that children work because parents are selfish, analysts have lost interest in the “nimble
fingers” hypothesis. It is almost certainly the case that if employers had to pay the true social
cost of employing children, they would find technological innovations to replace child workers.
In fact, Levison et al. (1996) and Anker and Barge (1998) find that children are not necessary for
the Indian carpet industry. Both authors argue that it is more likely that children are attractive
employees in spite of their low productivity because they are less aware of their rights and more
willing to take orders, do monotonous work, and are less likely to steal.
We can draw other bits evidence on the demand side as well. For example, Parsons and
Goldin (1989) note that the LFP rates for children did not vary across sectors, as would be
expected if children had some special characteristics that made them particularly valuable to
industry. First, a cross-state analysis of the U.S. 1900 Census of Population reveals that the LFP
rates for males 10-15 was not significantly higher in manufacturing and mining than in
22
agriculture, although, it should be noted that the LFP rate for girls aged 10-15 was higher in
industry. However, those industries in which one would expect the special features of children to
place them in high demand, such as textiles, boots and shoes, paper, and clothing, did not employ
substantially more children than iron, steel, and mining. Rather, the industry in which a child was
employed was overwhelmingly determined by the industry in which the parent was employed.
Concluding Remarks on the Determinants of Child Labor
1. Parental education plays a persistent and significant role in lowering the incidence of
child labor, above and beyond the impact on family income. The results presented on this are
quite robust, as reviewed by Strauss and Thomas (1995). In some cases, such as Cote d’Ivoire,
the parent’s level of education overwhelms all other family characteristics.
Several theoretical contributions on the determinants of child labor emphasize the
importance of educating a single generation of parents and its long-term implications for
decision-making for future generations. The theoretical mechanism draws attention to the impact
that an education has on the parent’s human capital and income. That is, educated parents earn
enough income to afford to educate their own children.
However, the empirical evidence very strongly suggests that a parent’s education affects
future generations above and beyond the impact on household wealth. There are several possible
explanations. For example, educated parents have a greater appreciation for the value of an
education or uneducated parents may simply want to believe that the human-capital decisions
made by their own parents were correct. In any event, cost-benefit analysis of programs that
concentrate on educational attainment must look beyond the impact that an education has on a
future parent’s income stream and incorporate the implications for human-capital formation by
subsequent generations.
2. Several studies point to the importance of school quality as an important determinant
of schooling and work. However, school quality is virtually never measured directly in any of the
23
studies discussed. It is quite possibly the case that when a family poised to move children out of
the work force and into school fails to do so, the culprit is poor schools. Poor school quality is
found to be weakly important in rural Ghana (Lavy, 1996) and very important for Africa
generally (Bonnet, 1993). It should be noted though that even if poor school quality lowers the
value of formal education, there is an abundance of empirical evidence across Latin America,
Africa and Asia that the return to education is still quite high and more than offsets the foregone
income of children in school.
3. The role of household income in determining child-labor decisions needs further
study. Clearly, there is a very strong cross-country negative correlation between child labor and
per capita GDP. However, the role of family income is not so predominant in explaining
variations within a community. We did observe, for some but not all countries, that household
expenditures play a central role in child labor decisions. This evidence suggests that there are
some external effects across families that make it difficult to put children in school even as
income rises or, equally, difficult to put children to work when income is critically low. In
particular, none of the studies does a very good job of measuring school quality. The role of cost
of schooling, when this is measured, suggests that it may be acting as a proxy for quality. In this
case, parents who have the financial ability to forgo the income from the children may still not
choose schooling if the quality of schools is very poor.
It is also the case that regional dummies and cultural characteristics such as religion or
caste have some significance, suggesting a nontrivial role for cultural factors.
Finally, it is important to reconsider when economic theory tells us to expect poor parents
to put their children to work. Recall that, as argued by Baland and Robinson, child labor is a
device for transferring resources from the future into the present. Children who work do not
invest in human capital that will make them more productive in the future. A family will choose
to make this inter-temporal shift in household resources when current income is low relative to
future income. Thus, it is not the absolute level of family income that matters for the child labor
24
decision but, rather, the current level relative to future income. There may be families that are
quite poor and do not have any reason to expect any change in the future. Such families have no
reason to attempt to consumption-smooth by putting their children to work.
4. Household assets play an important role in the child-labor decision. One might
expect that the more assets a family has, the lower the probability of child labor. However, there
are a number of assets that require a complementary input of labor, and families may expect to
get that labor from their children. Tapping the human capital of mothers in the family also
requires an increase in child-labor inputs in home production. Thus, a strategy of increasing
access to capital markets may not always lower child labor, at least in the short run.
Nevertheless, the significant role of household assets lends some evidence to the
possibility that incomplete credit markets give rise to inefficiently high levels of child labor. For
example, the presence of older children in the home considerably reduces the probability of child
labor. Note that there is a measurable impact above and beyond the contribution that older
siblings make to family income. This is particularly the case for older brothers, who embody the
greatest human capital. In addition, a parent’s education reduces child labor for reasons other
than the impact of education on the parent’s productivity. It is possible that a parent’s education
is viewed as a marketable asset, or it may be a reflection of the informational externalities
associated with the value of formal education.
What is not clear is why family assets matter. On the one hand, households with assets
can more readily weather adverse events. That is, these assets provide the household with the
ability to manage uncertainty and, as a consequence, child labor is not required for this purpose.
However, families with assets may also have more access to capital markets or can, themselves,
fund a child’s education without a formal loan. That is, household assets help families transfer
household income intertemporally.
In either case, expanding access to formal capital markets to families who otherwise lack
collateral may lead to a reduction in LFP rates for children. However, it is also the case that
25
placing constraints on household decision-making, such as mandatory schooling, may at least
inhibit the family from turning to internal assets that can be accessed only if children work more.
Providing working mothers with firm-level child care may also help reduce the reliance on older
daughters to care for their younger siblings.
5. It is clear that older children are more likely to work than younger children. As
children grow older and acquire skills, the opportunity cost of schooling rises. This is particularly
the case for adolescent boys, who are increasingly able to perform physically demanding tasks as
they approach maturity. Thus, it appears to be the case that it will be more challenging and costly
from a policy point of view to induce older male children to remain in school.
6. The role of siblings in the household does not appear to be a major deterrent to
schooling once we control for other household characteristics. The only exception is that there is
some evidence that in some cases, mid-aged children are caring for younger siblings.
When evidence that older children are caring for younger children is combined with the
fact that the presence of an older sibling in the house generally raises the probability of schooling,
it is possible to make a case that parents are diversifying their human-capital investments in their
child assets. The oldest children acquire human capital in the form of on-the-job training and the
youngest children receive formal education. However, this interpretation of the evidence does
not accord well with the other significant result: the presence of siblings in the same age range
tends to raise the probability of school and lower the probability of work.
Rather it seems more natural to, first, view children in the middle age range, 10-14, as
complements, sharing housework and schooling opportunities. Second, when we observe older
children making schooling possible for their younger siblings, this is likely evidence that older
siblings help relax the liquidity constraint in the presence of capital-market failure. Third, when
we observe mid-aged children caring for younger siblings, it is to help the family make optimal
use of the mother’s human capital in the form of marketable skills. Thus, policies that focus on
lowering fertility may not be particularly effective in reducing child labor.
26
To the extent that parents diversify their child assets, this appears to occur along gender
lines. In Latin America, parents are more likely to engage in the formal schooling of their
daughters whereas in Africa parents are more likely to school their sons.
7. Finally, we see little evidence that child labor is driven by the needs of industry.
Children are far more likely to be working in a rural setting rather than an urban setting where
factories are located. In addition, LFP rates rise with a child’s age, strongly suggesting that the
productivity of a child increases, the larger and stronger the child is. If child workers were valued
for their small stature and tiny fingers, we should have observed the opposite. To the extent that
child labor is a demand-side phenomenon, it appears to occur primarily within the household.
Families with a household enterprise or a large tract of land tend to want to put their children to
work. That is, the household’s physical assets are most efficiently employed when the child’s
time is used as a complementary input.
IV.
Traditional Policies Targeting Child Labor and Education
We now turn to an examination of some traditional strategies for reducing child labor and
increasing school attendance. As discussed in Engerman (2001) in this colloquium, the initial
strategies for circumscribing child labor in industrial England included limits on hazardous work,
hours of work, minimum age of work, a prohibition against night work and minimum educational
attainment for working children, and strategies to promote general economic growth.
Promotion of Economic Growth
Given the strong correlation between economic growth and the decline in child labor,
some have argued that policies targeted explicitly on child labor are ill-conceived. There is
certainly an abundance of evidence both that household income is an important determinant of
whether and how much children work in developing countries, as well as a strong negative
correlation between per capita GDP and income growth both across countries and across time.
27
However, while economic growth undoubtedly offers the greatest promise for helping children in
the long run, the well-being of today’s children and perhaps even economic growth itself may
depend on getting children out of jobs and into schools today.
Minimum Age and Compulsory Education
In the international arena, child-labor practices are also regulated by ILO Convention 29
that calls for the abolition of forced labor and Convention 138 that provides for a minimum age of
employment. According to the language of Convention 138, children should not enter the labor
market before having completed compulsory education or having reached the age of 15.
Additional provisions allow for light work beginning at age 13 and hazardous work beginning at
age 18. Furthermore, developing countries may permit light work at the age of 12 and nonhazardous work at age 14. Indeed, regulating minimum age and compulsory education has
become the most common strategy globally for limiting child labor. Typically, children are
required to attend school until the age of between 14 and 16 and are permitted to begin working at
the age of 14.
There are several theoretical justifications for compulsory schooling and minimum age of
work laws. Certainly if there is evidence that parents are overly selfish when making decisions
concerning investing in the human capital of their children or if there is an external effect of
education of the sort argued by Grootaert and Kanbur (1995), then requiring parents to provide
for more education than they would freely choose can be justified on both equity and efficiency
grounds. We might also be able to justify constraining family choice if parents are making
school/work decisions to diversify the human-capital investment in their children. It may be
legitimate to prevent parents from schooling some of their children and putting others to work on
equity grounds if on-the-job training is less valuable than formal schooling once the child grows
to adulthood.
28
In addition, Basu and Van’s (1998) multiple-equilibrium argument in support of a ban on
child labor can be implemented by establishing laws regulating minimum age of work with
monitoring by examining school-attendance records. Finally, to the extent that labor standards in
the international arena require countries to control child labor, presumably all countries
attempting to be in compliance with international standards would pass and attempt to enforce
compulsory schooling and minimum age-of-work laws.
Minimum age of employment and years of compulsory education are reported for a
selection of countries in Table 5. We also report labor force participation rates for children
covered by the legislation. Needless to say, many families in developing countries do not comply
with the law. In Latin America, the LFP rate for children aged 5-14 is 17 percent even though in
nearly all countries children are not legally permitted to work until the age of 14. A similar
situation exists throughout Asia, in which the LFP rate for children aged 5-15 is 21 percent. In
Africa, over 40 percent of 5-14 year olds are working even though the minimum age of
employment is typically 14 years or even higher. Although these age ranges overlap somewhat
with the ages of legal work, these numbers nonetheless indicate a good deal of illegal child
employment.
Similarly, Krueger (1996) presents evidence from the 1990-91 waves of the World
Values Survey2 in which respondents are asked for the age at which they completed (or will
complete) full-time education. His results are reported in Table 6 for individuals born between
1959 and 1974 for a select group of countries. These results clearly indicate that while
compliance is the norm in high-income countries, compulsory-education laws have little
discernible effect in low-income countries.
In the United Kingdom, the law was changed in 1947, raising the age from 14 to 15, and
then again in 1973, raising the age from 15 to 16. For each cohort, the modal age at which
2
The World Values Survey is undertaken by the European Values Systems Study Group.
29
children left school coincided with the legal requirement and no more than 5 percent of children
left school before the legal age.
However, when the law is somehow inconsistent with the equilibrium level of education,
there is little effect. For example, Brazil increased the compulsory school age from 11 to 14 in
1971. Yet 85 percent of children still left the work force before reaching the age of 14 whether or
not they were covered by the revised regulation.
Furthermore, the evidence of compliance is corroborated with evidence from earnings, at
least in the United States in the period 1960-1980. Angrist and Krueger (1991) and Harmon and
Walker (1995) find that the earnings payoff to years of compulsory school is higher than for years
of optional schooling.
Clearly, the casual empirical evidence does not suggest that laws regulating compulsory
education and minimum age of work are very effective in controlling child labor in those settings
in which child labor is problematic. In order to gain a sense of how laws regulating minimum age
of work and compulsory education might operate in a developing country, it is instructive to
analyze the effects of such laws during the period in which child labor was declining rapidly in
industrialized countries.
Several studies look at the historical events surrounding the decline in child labor in
Western Europe and North America throughout the 19th and into the 20th centuries. Scholliers
(1995) studied child labor in Ghent, Belgium and concluded that the incidence of working
children under the age of 12 declined substantially by the middle of the 19th century without legal
intervention. Brown, Christiansen and Philips (1992) draw similar conclusions for the U.S. fruit
and vegetable canning industry between 1880 and 1920. While legislation played some role,
economic forces dominated the decision to remove children from this sector. By contrast, BolinHort (1989) argues that legal restrictions played a substantial role in the removal of child workers
from the cotton mills in Manchester, England.
30
Thus, it is useful to consider some of the more careful statistical analysis of the impact of
laws regulating entrance to the labor market and compulsory schooling. Angrist and Krueger
(1991) develop a “natural experiment” type statistical technique for evaluating the impact of
compulsory schooling laws on school attendance. The 1960-1980 U.S. censuses collected
information on the “quarter of birth” and “school attendance as of April 1.” Angrist and Krueger
argue that if compulsory school laws are effective, teenagers who are 16 years old as of April 1
and live in a state that requires students to remain in school only until they are 16 are less likely to
be attending school at the time of the census than 16 year old teenagers who live in states with a
mandatory school age of 17 or 18. They find a statistically significant effect of compulsory
schooling laws for 1960 and 1970, thus supporting the hypothesis that laws affect schooling
behavior.
Acemoglu and Angrist (1999) perform similar analysis on the same data looking for the
impact of child labor laws on school attendance. They find, for example, that boys born in states
that required 9 years of school before entering the work force spent 0.26 more years in school
than boys born in states requiring 6 or fewer years of schooling.
The Angrist-Krueger technique was then applied to earlier periods in U.S. history.
Margo and Finegan (1996) analyze the schooling choices of 14 year-olds reported in the1900
federal census. In this study, 14 year-olds are broken into two groups: (1) those teenagers who
are already 14 at the beginning of the 1900 school year; and (2) those who are not yet 14 at the
beginning of the school year. Margo and Finegan hypothesize that if mandatory school laws are
effective, the younger 14 year-olds living in a state with a compulsory schooling law should be
more likely to be in school than older 14 year-olds. However, no such difference should exist for
14 year-olds in states without compulsory schooling laws. Margo and Finegan find that
compulsory school laws have a positive and statistically significant impact on the decision to
obtain some schooling for younger 14 year-olds. However, the laws have no discernible effect on
the probability of full-time school attendance.
31
They then consider the impact of compulsory school laws combined with laws that
regulate the minimum age of work. The addition of child-labor restrictions is likely to have an
additional effect on school attendance because child labor laws were more aggressively enforced
than mandatory education laws at that time. In this case, the combination of laws has a
statistically significant impact on school choice. Young 14 year-olds were 18 to 21 percent more
likely to obtain some schooling if their access to the labor market was legally restricted.
However, the laws did not significantly increase the probability of being in school full-time.
The statistical evidence presented here has been criticized, most notably by Moehling
(1999). She argues that the laws mandating school attendance are, themselves, endogenous and
tend to follow the decline in child labor rather than precipitate it. That is, cross-state differences
in technology, immigration and real wages are driving both the change in educational attainment
and the laws regulating school attendance. Thus, despite the fact that compulsory education laws,
child labor laws and school attendance are correlated, it is not a causal relationship.
Moehling looks at occupation rates – the proportions of youth who identify some form of
employment as their main use of time, as opposed to school. Then, in order to control for
differences in the economic conditions across states that might be driving both the legislative
process and schooling choice, she first looks at the difference in occupation rates for 13 and 14
year-olds in each state prior to the introduction of compulsory schooling laws. This gives a
baseline against which to compare the difference in occupation rates for 13 and 14 year-olds after
some states passed compulsory education laws. Moehling also included a number of other
economic and demographic variables that have been shown to play a significant role in child
labor decisions, as discussed above.
Moehling finds that the probability a 14 year-old boy would be working fell substantially
between 1890 and 1900 in states with newly enacted compulsory education laws. However, she
observes a statistically similar decline in labor force participation in states without such laws.
Similarly, the labor force participation rates for 13 and 14 year-old girls in states that did pass
32
compulsory education laws also fell between 1880 and 1990. By contrast, 13 and 14 year-old
girls in states that did not pass compulsory education laws had increased labor-force participation
(LFP) during the decade. Thus, for girls, there is a negative correlation between the passage of
laws and the LFP rates for girls.
However, there is no differential effect on girls covered and not covered by the law. That
is, the employment choices by 13 year-old girls covered by compulsory education laws is
mirrored by 14 year-old girls in the same state not covered by the law. From this, Moehling
infers that the failure of some states to pass laws requiring 13 year old-girls to attend school, and
the increase in the employment of 13 year-old girls in these same states, are being simultaneously
driven by other economic factors.
Moehling then goes on to consider Margo and Finegan’s hypothesis that schooling is
affected by the combination of child labor and compulsory education laws. Once again, the laws
do not seem to be driving behavior. The only case in which 13 year-olds behave differently than
14 year-olds occurs for boys in states with no legislative change. In states with no laws
regulating either compulsory education or minimum age of employment, the LFP rate for 14
year-old boys rose between 1890 and 1900, whereas the LFP rate for 13 year-old boys declined
during the same period. Thus, the results are running precisely counter to the expectation that
laws affect behavior!
In response to the rising LFP rates for girls in the last decade of the 19th century, there
was a burst of legislative activity shortly after 1900. In 1900, 24 states had laws regulating
minimum age of employment. By 1910, 43 states had such laws. Perhaps more importantly, by
1909, 34 states had enacted legislation providing for inspectors assigned to enforce child labor
laws.
Moehling then applied her statistical technique to the 1900 and 1910 censuses. In this
case, the estimated effect of legal restrictions on school attendance, at least, appears to be positive
33
but statistically insignificant for some groups. However, the impact is small relative to the timeseries change.
What can we conclude from this evidence? First, the more carefully executed the
statistical analysis, the weaker is the evident effect of legal restrictions on child schooling and
labor decisions. Second, it does appear that for carefully crafted laws, such as those enacted in
the last quarter of the 20th century in England, there is some impact of legislation on behavior at
the margin. However, when the age limits specified by the laws are substantially at odds with
optimizing decisions by households, they have little effect. For example, the laws written in the
United States around 1900 tended to specify 14 years as the cut-off between schooling and work.
However, Moehling’s evidence clearly suggests that 14 years of age was not viewed as a
significant work-school boundary for many U.S. households at that time. Similarly, recently
enacted laws regulating work in Brazil have had no effect on household decisions. Thus
mandatory school laws and minimum age of employment are at best, a complement to other
policies designed to alter the family’s perception of the appropriate age at which children should
begin working.
Finally, the results of Margo and Finegan on the one hand, and Moehling, on the other,
are not as inconsistent as they may at first seem. Margo and Finegan focus on the 1900 census
because it not only asks whether a child views school as the main occupation but also whether the
child is in school at all. Moehling, by contrast, looks at several decades of data and, so, is only
able consider whether the laws are affecting a child’s perception as to whether school is the main
occupation. Neither study finds an impact of compulsory schooling laws or child-labor laws on
the child’s perception of his/her main occupation. That is, neither study finds that the legal
restrictions increase the probability of full-time schooling.
34
V.
Recent Policy Initiatives in the Addressing Child Labor
Although the empirical results discussed in Section III are by no means conclusive, they
are certainly suggestive of the types of policies that might be effective in reducing the incidence
of child labor. We turn now to consider some of the policy initiatives that have recently been
undertaken in some developing countries.
Recently, several governments have implemented a range of positive strategies designed
to improve compliance. For a more thorough description see U.S. DOL (1998) and World Bank
(2001), from which much of this section draws. See also Anker and Melkas (1996) for an
overview of programs relying on economic incentives, worldwide. The programs include
improvements in educational infrastructure, programs targeted at children who have fallen or are
likely to fall behind in school, financial incentives and sector-specific programs. We discuss
some of these below.
Educational Infrastructure
Increased spending on books, supplies, buildings, and teacher training have been pursued
by several governments. Brazil has been one of the most aggressive in this regard. Beginning in
1997, the Livro Didactico project has provided $142.5 million for textbooks. The TV program
TV Escola is targeted at raising the skill levels of teachers in rural areas. The program also
includes the distribution of kits that contain instructional materials. Funds have also been made
available to raise the wages of extremely low paid teachers and to build and improve publicschool facilities.
The Mexican government also uses telecommunications to improve education quality for
rural communities. By virtue of the Telesecundaria program, rural seventh, eighth and ninth
graders can view educational programs broadcast by the Mexican Ministry of Education. The
central government provides a teacher, television set, satellite dish, decoder, instructional material
and books for qualified schools.
35
Some poorer countries have had to rely on the one-room schoolhouse model in order to
extend educational opportunities to all children. For example, The Ministry of Education in
Egypt built 8,500 new one-classroom schools in rural communities during the mid-1990s and
increased investment in teacher training. Similarly, the government of the Philippines established
1880 “incomplete” elementary schools, along with 900 elementary schools, thereby halving the
proportion of barangays (political divisions of municipalities or cities) without a primary school.
The Turkish Ministry of Education has built 670 new primary schools and appointed
1,930 new teachers in order to implement a new compulsory school law requiring eight years of
schooling. Nevertheless, many communities in Turkey still lack most of the items essential to a
school, such as chalk, blackboards, teachers, books, etc.
Remedial Teaching and Flexible Schedules
Working children, given the competing work and school demands on their time, are
particularly likely to fail to complete each grade with their cohort. Some empirical evidence
discussed above suggests that greater flexibility in school schedules would help working children
remain in school. Nicaragua’s remedial education program, Extra Edad, targets older children
who have failed to complete the primary grades by the age of 14. Classes are offered after work
in order to allow the child to continue to earn an income while pursuing an elementary education.
Guatemala has also introduced a strategy of flexible schedules to keep working children in
school. Classes begin after market work is completed and students make up missed schoolwork
with independent study. Children of migrant workers are also offered a more flexible school
calendar, allowing students to resume school attendance as soon as they are able. Mexico
provides flexibility by allowing the children of migrants to attend school in whichever district
they happen to be currently residing. Peru offers classes in three shifts throughout the day. This
school schedule allows each student to combine work and school in a manner consistent with the
requirements of the employer. Lapu-Lapu City in the Philippines offers a work-study program in
36
which children attend school in the morning and then report for work in the afternoon. Child
workers are directed toward less dangerous work and monitoring of working conditions is
intensified in hazardous occupations such as firecracker assembly.
The state of Andhra Pradesh in India launched a program in 1997 targeting children who
have left school or were never enrolled with two-month school camps. Typically eligible
children are bonded child laborers, domestic servants and those from lower castes. Camps are
comprised of 100 children and 5 teachers each. The pilot program was particularly successful as
a stepping-stone to formal education. Of those children enrolled in the first year of the program,
74 percent subsequently enrolled in formal school. The Andhra Pradesh program is particularly
attractive in light of the fact that 60 percent of children aged five to 14 never attend school.
Financial Incentives
Governments rely on a wide array of financial incentives either to make school more
attractive or even to make school attendance feasible for families. Among the most popular are
school meal programs. Such aid programs are distinctive because they tie the aid to school
attendance. Brazil launched Marenda Escolar in 1997, spending $453.4 million on breakfast and
lunch. Urban Brazilian families who are likely to put their children to work also receive food
baskets from the Foundation of Childhood and Adolescence. Like food distributed at school, the
food baskets are contingent on school attendance. Mexico provides approximately 4 million
breakfasts a day to poor children attending school. All Egyptian children are also given one
meal/day in school. Similarly, the government of South Africa provides meals for five million
children who attend school.
While food aid may make school more attractive, it may not be a sufficiently strong
incentive to induce families to give up the income earned by children. As a consequence, some
governments have instituted cash stipends or in-kind gifts for children attending schools. For
example, the poor families in Bangladesh receive 15 to 20 kg of wheat per month if their children
37
are attending school. In 1996, the program reached 1.14 million families. ILO-IPEC (1998a)
finds that the program has a significant impact on enrollment, attendance and drop-out rates.
In Brazil, Bolsa Escola pays a monthly stipend to each family with an unemployed adult
in the Federal District that keeps all of its children aged 7 to 14 attending school. In addition, the
program makes a deposit equivalent to one month’s salary into a savings account after each year
of completed school through the eleventh grade. Funds are forfeited if the child fails to advance
to the next grade.
Mexico introduced a similar, though not identical, program in 1997. The Program for
Education, Health and Nutrition (PROGRESA) targets over 2.5 million families whose children
are not attending school. The program pays a bimonthly stipend to the families of children who
maintain an 85 percent attendance record. The stipend ranges from $7 to $63, depending on the
age and gender of the child. The program also provides families with funds to purchase school
materials and supplies, a basic package of primary health-care services, and food supplements for
children and mothers. The health-care provisions of the program are tied to routine visits to
medical facilities.
Although the Mexican and Brazilian programs appear similar, some key aspects are
likely to make the Mexican program more effective at lowering child labor. The Brazilian
subsidy to families with an unemployed adult has features that repair some of the effects of
capital-market failure. In the absence of the program, families without access to capital markets
are forced to turn to the labor of their children in order to survive periods of economic adversity,
such as an unemployed adult. Thus, this program must be seen primarily as a program to deter
child labor that occurs as a form of family insurance against income uncertainty. Children who
work as a consequence of poverty proper may not be affected.
The educational savings account is even more deceptive. The family can access the
account only after the child has successfully completed eleven years of education. Therefore, the
savings account cannot serve as collateral for education loans, nor can the family access the
38
account to pay ongoing expenses. In addition, the child cannot pre-commit to surrender the
proceeds to repay their parents even if the parent could access the capital markets on behalf of the
child. As a consequence, none of the problems with capital market failure are remedied with the
Brazilian savings-account program.
The only impact the loan has is to raise the present discounted value of an education
relative to current income. The increased return to education may affect the calculus in a family
that is able to borrow in order attend school, but it cannot help families without access to capital
markets.
By contrast, the Mexican program buys out the labor contract of the child from the
parent. Participating children receive a stipend that partially replaces the income the child could
earn by working in exchange for school attendance. Thus, all of the problems with capital-market
failure and their implications for inefficient child work are sidestepped. Issues of collateral and
intra-family bargaining are no longer relevant. Nor do policy-makers need to be concerned that
providing access to capital markets will lead families to purchase assets that they intend to
combine with the labor of their children.
Subsidy programs that replace the child’s income boast some of the highest success rates.
Between 1995 and 1996, the official dropout rate in the Brazilian Federal District fell from 11
percent to 0.4 percent, although the extent to which this should be attributed to any particular
program is unknown. Similarly remarkable success is reported for the Brazilian Child Labor
Eradication Program (PETI). Begun in 1996, PETI targets nearly 900,000 Brazilian children
aged 7 to 14 working in the most harmful conditions in rural areas. Under the program, mothers
in families earning half the minimum wage per capita receive a monthly stipend equal to
US$13.50 per month for each child attending school and after-school programs full-time.
Children in school also receive three meals per day. An equal amount of money per child in the
program is paid to the local municipality to finance salaries, materials and meals. In return, the
municipality must pay 10 percent of the cost of the schools’ infrastructure. Monitoring of school
39
attendance is undertaken by teachers. The total cost of the program through 2006 is estimated to
be on the order of $2 billion.
Anecdotal evidence suggests that PETI is profoundly successful. The town of Conceicao
do Coite, a Brazilian sisal-producing community located in Bahia, provides a striking example.
Children working in sisal harvesting are commonly permanently injured by both the sharp sisal
stalks and the tools used for cutting. Evaluation of PETI undertaken by UNICEF suggests that
child labor in Conceicao do Coite has been virtually eliminated.
One of the distinctive features of the PETI program is that it combines stipends to
families that replace the child’s earnings with financial support to develop and fund educational
opportunities, all of which are embedded in a vigorously active local community committed to
eradicating child labor. In addition, PETI, like Bolsa Escola, is a means-tested program targeting
the very poor. Finally, the educational subsidy is quite large in comparison to the family’s
income. Although the size of the benefit and the income cut-off have varied over the life of the
program, the educational subsidy has in some cases been equal to the income earned by the
parents.
However, it is difficult to judge the quality of the reported evidence. The teacher’s report
on school attendance is required to receive the subsidy and also serves as the basis for schoolattendance statistics. Teachers may have an incentive to misreport, either because of bribes or
concern for the welfare of the child.
Both Brazil and Mexico have designed additional income-support programs targeted at
specific sectors. The Mexican government targets children working in the fruit and vegetable
sector in the northeastern state of Sinaloa. Aid is paid in the form of food packages worth about
30 percent of an adult’s monthly salary. As with the income supplement, families are required to
demonstrate a substantial school-attendance record of their children. Local growers are required
to contribute 30 percent of the cost of the food. Growers may also construct and furnish local
schools. In such cases, the government provides teachers and supplies.
40
The ILO’s International Program on the Elimination of Child Labor (IPEC) furnished
seed funding to start a program sponsored by the Union of Rural Workers in Retirolandia in
Brazil to provide families with assets that they could use to support their children in school rather
than send them to work. The Goat-to-School program provides each eligible family with a goat
and information on tending and rearing goats. Beneficiary families are required to use the milk to
feed their children and to repay the program in goats without interest. This unique program
provides families with the assets they need to find safer alternative employment for their children
that does not interfere with schooling.
The Goats-to-School program is not very significant, however, in terms of the number of
children covered. Between 1996, when the program was begun, and 1998, 60 goats were
distributed to 30 families affecting 100 children. The incentives and constraints built into the
program are quite sensitive and responsive to the evidence currently available as to why children
work and in what occupations. Clearly the program provides families with an asset that produces
an income stream that the family can rely on rather than on the labor services of their children.
That is, poor families are able to acquire capital that allows them to fund current education for
their children, thus eliminating inefficient child labor associated with incomplete capital markets.
In addition, the loan can be repaid through the efforts of the children tending the goats since the
loan is repaid simply by returning one baby goat to the program for each adult goat received.
Thus, the intra-family bargaining problem that arises because children cannot pre-commit to
repay loans taken out on their behalf is eliminated because the children, through their efforts
tending the animals, are able to repay the loan on their own.
Furthermore, the children tend the goats, thereby continuing to make some current
contribution to the family. However, the times at which the goats need tending do not conflict
with schooling, thus providing each child with sufficient flexibility in their work schedule to
combine school and work. Nor is the work so onerous that the children are too exhausted to
complete their schoolwork. Finally, receiving the benefit is contingent upon school attendance.
41
As a consequence, the program provides a strong incentive to substitute education for work even
if the family is far from the income level that would place them near the work-school margin.
Thus, it is not necessary to wait until income reaches some critical level at which parents start
withdrawing children from school and the implicit subsidy does not have to be so large as to raise
income to the poverty line to be effective.
To the extent that Goats-to-School has a design weakness, it is the absence of time
consistency. Families receive the asset based on a commitment to place their children in school.
However, there is no mechanism for enforcing ongoing compliance other than the social pressure
that might be brought to bear by the union implementing the program. The income subsidies
described above that make a payment only after the teacher certifies attendance may prove to be
more effective in lowering the level of child labor for a given level of expenditure.
Another interesting feature of the Goats-to-School program is that it is self-sustaining.
Animals repaid become assets for new families entering the program. Although IPEC provided
the original funding, the program is now self-financing.
Some programs that provide financial support are specifically targeted at replacing the
contribution that the working children make to household income. Others are targeted at helping
families defray the cost of education. For example, the Egyptian government pays a grant equal
in value to about US$4.17 to cover uniforms, books and supplies for families earning less than
about US$29.41 per month.
Programs to Reduce Child Labor in Targeted Sectors
Conditions for working children in some sectors are sufficiently hazardous that programs
have been tailored to the specifics of the relevant sector. Cultural factors may be sufficiently
complex that simply relying on financial incentives may be ineffective. Examples include the
Vale dos Sinos Project initiated in 1996 to reduce the employment of children in the Brazilian
footwear industry and the HABITAT project initiated in 1998 to reduce child labor in the stone
42
quarries of Guatemala. Both projects have a public-education component designed to sensitize
parents, employers and the community to the risks to children employed in these sectors.
Program objectives also include improved working conditions, medical services and flexible
school options. The government of Peru has also targeted children who work turning bricks in
the Huachipa brick fields outside of Lima. In addition to mentoring and tutoring young children,
the program provides health care and small business loans to start a family enterprise.
Providing alternative employment opportunities has also been used as a strategy to draw
children out of the quarries of Carabayllo, Peru. Mothers who agree to keep their children out of
work receive financial and other help in establishing a micro-enterprise making plastic bags.
Families are provided with equipment, raw materials and technical advice on beginning the
business.
A similar program has been developed in Turkey. It is a common practice in the
mountain villages of the Duragan district of Sinop to auction off male children aged nine to 15 to
help during the harvest season on the farms of affluent families. The Development Foundation of
Turkey has launched a program to train families in small-scale agricultural projects, such as beekeeping and turkey-breeding, that allows the child to work productively while remaining at home.
Children can, in some cases, earn more in these newly-created family enterprises than as rented
labor.
Several programs are targeted at raising awareness of the negative effects of work on
children. For example, the African Network for the Prevention and Protection Against Child
Abuse and Neglect on the tobacco and tea plantations in Tanzania uses drama and theatre to
mobilize communities and educate. Teachers report increased attendance and some employers
have begun to provide financial help to schools for the purchase of supplies and school meals.
Children who have been formerly bonded frequently perform poorly in a formal
education setting. In Nepal, rehabilitation is undertaken by the Informal Sector Service Center,
43
providing nine months of remedial training in language and arithmetic. Children are then
channeled into elementary schools.
Several communities have reached formal agreements with employers to not employ
children and to return currently employed children to school. For example, the Bangladesh
Garment Manufacturers Association (BGMEA) signed a Memorandum of Understanding in
1995, which provides for removing children currently working if they can be placed in school. In
addition, no under-age children should be newly employed. Children presenting themselves for
employment shall be directed to NGO-run schools where they can receive a monthly stipend
equal in value to about US$6.88 for attending school.
The program appears to have been fairly successful. In 1995, 10,546 children were
working in BGMEA factories. About 43 percent of member factories employed children. This
figure fell to 32 percent in 1997 and 13 percent in 1998 (ILO-IPEC, 1998).
Several U.S. importers of soccer balls have signed the Partners’ Agreement to Eliminate
Child Labor in the Soccer Ball Industry in Sialkot, Pakistan. The program aims to provide
children removed from employment and their younger siblings with informal education,
alternative income-generating opportunities, formal schooling and awareness training for parents.
The weakness of this program occurs in the monitoring component. It is commonly the
case in Sialkot for women and their children to stitch soccer balls in between other household
chores. In order to prevent families from putting their children to work stitching soccer balls,
work has been moved from homes to stitching centers. However, as has been noted in a previous
section, mothers who work outside of the home place their daughters at risk for full-time homework. However, when mothers work in a household enterprise, such as soccer ball stitching,
daughters can more readily combine home-work with schooling. As a consequence, this program
has the potential to undermine the efforts that Pakistani families are making to educate their
daughters.
44
A rescue and rehabilitation program was initiated in 1995 by the National Society for
Protection of Environment and Children in the Nepalese carpet industry. The program attempts
to provide informal education for children removed from work but also conducts classes for
children in the carpet factories. Children removed from work are either returned to their family
or placed in youth hostels where they receive other skills training.
Empirical Evidence on Program Effectiveness
Although programs that provide incentives to replace work with schooling seem like they
ought to be effective, there is very little careful empirical analysis of these programs. Anker and
Melkas (1996) surveyed administrators of 68 income-replacement and substitution activities in
Africa, Asia, and Latin America. Respondents generally thought that their programs were
effective. However, the authors doubted the value of such self-reporting and noted that none of
the programs had evaluated the impact on the incidence of child labor either short-term or longterm.
Nevertheless, the survey provided some useful insights.
1. A package of school-based incentives, remedial education, income-generating
activities for families, and awareness training for parents is more effective than any
one of these components individually.
2. School-lunch programs themselves do not provide a sufficient incentive to draw
children out of work and into school. As a consequence of the low financial value of
the meal combined with the poor quality of schools, school-lunch programs do not
generally alter the parents’ calculation of the value of school relative to work.
Children covered by the survey typically contributed 20-25% of the family’s income,
an amount that is far in excess of the value of a single mid-day meal.
3. Many respondents were concerned that cash grants could be misused and so preferred
aid in-kind. However, it is also the case that cash provides families with flexibility
45
that is sometimes essential to the effectiveness of the program. Most respondents
preferred programs that had elements of both.
4. Programs that provided apprenticeships, school-work combinations, or “safe work”
alternatives and other informal education were particularly effective in helping
children leave dangerous or onerous occupations. This is partly a reflection of the
very poor quality of schools that families frequently regard as irrelevant to their
situation.
5. Several respondents raised concern with dependency and the role that incentives
might play in luring children into work in order to qualify for benefits.
Although the survey evidence cited above is of some value, some programs have been
carefully evaluated using statistical techniques. Ravaillion and Wodon (1999) evaluate the Foodfor-Education (FFE) program in rural Bangladesh. Participating households receive a food ration
of rice as long as their children attend school. In 1995-96, 2.2 million Bangladeshi children
benefited from the FFE program. The national government targets economically underdeveloped
areas for benefits. Local community groups then select participants based on idiosyncratic local
information. Children are required to maintain an 85% attendance rate. Monitoring and food
distribution are handled by each school’s headmaster.
Based on the 1995-96 Household Expenditure Survey undertaken by the Bangladesh
Bureau of Statistics, the typical participating household received 114 kg of rice per year. Using
information on the local cost of rice, average family size, and local wages, Ravaillion and Wodon
calculate that the FFE stipend is the equivalent of 13 percent of the average monthly earnings of
boys and 20 percent of that for girls.
Ravaillion and Wodon estimate the determinants of the probability of work and school.
A working child is one who regarded work as his/her “normal activity” in the previous week.
The level of education is measured by the reported educational status for children 5 to 16 who
46
have not completed primary school. Explanatory variables include distance to school, the type of
school, school-quality variables, parental education, community religion, household
demographics, land ownership, the child’s age, and the size of the FFE stipend.
Ravaillon and Wodon find that the stipend has a strong and statistically significant impact
on both the probability of work and the probability of schooling. In particular, they find that a
FFE stipend of 100 kilos of rice increases the probability that a boy will be in school by 17
percent and that a girl will be in school by 16 percent. Thus, there is some evidence from
household survey data that corroborates the enrollment data provided by school administrators on
the impact of education subsidies on school attendance.
However, the impact on child labor is much smaller. An extra 100 kg of rice lowers the
probability of working as the main activity by 4 percent for boys and only 2 percent for girls.
Therefore, of the children newly in school, only a quarter of the boys and one-eighth of the girls
are switching from work as the primary activity to school as the primary activity.
Of course, it may be the case that some children newly in school were formerly working
only part-time, in which case the subsidy has had a positive impact on school attendance. Thus,
the subsidy appears primarily to increase school attendance at the expense of the child’s leisure
rather than work. These results strongly suggest that school quality or a failure to appreciate the
value of school, rather than each family’s need for their child’s income, are deterring formal
schooling. By contrast, for those children who are making a financial contribution to the family,
a stipend that replaces less than 20 percent of the child’s earnings is not sufficient to alter the
family’s calculation of the value of school relative to work.
Children nevertheless benefit from the program. On average, the total impact of the
subsidy raises family income. The average loss in child wages as a consequence of the subsidy is
only 19 percent of the average value of the subsidy. Furthermore, Wodon (1999) finds that
completing primary school in rural Bangladesh increases per capita consumption by nine percent.
47
Nevertheless, these results point more to school quality, an under-appreciation of the
value of education, or direct school costs as the most important deterrents to schooling, rather
than poverty and child labor.
Similar evidence has emerged for Thailand. Tzannatos (1996) finds that Thai children
under the age of 12 do not initially leave school in order to work. Rather the direct cost of
schooling, such as uniforms and supplies, relative to its value deters parents from keeping
children in school. Uneducated Thai children do not begin to enter the labor force until they are
12 to 15 years old.
The importance of the physical presence of school buildings in a community in raising
each family’s perception of the value of education is further supported by Duflo (2000). Between
1973 and 1978, the Indonesian Government built over 61,000 primary schools at a cost of US$5
billion. She finds that children aged 2 to 6 in 1974 received 0.12 to 0.19 more years of schooling
for each school constructed per 1,000 children. Duflo also finds a measurable impact on wages.
Each school built per 1,000 children also raised wages by 1.5 to 2.7 percent.
PROGRESA, a school subsidy program operated in Mexico, was specifically
implemented with the purpose of providing data that can be used to rigorously analyze the
program’s effectiveness. Design and implementation of the analysis was executed by the
International Food Policy Research Institute (IFPRI, 2000). At the inception of the program in
1997, households in seven states3 were randomly assigned to control and treatment groups. Of
the 506 communities initially targeted, 320 were designated for treatment and 186 as control
communities.
Before the program was implemented, PROGRESA surveyed rural households in the
targeted states in order to determine their socio-economic status. Employing the program’s
selection criteria, 78 percent of households were deemed eligible for benefits. Households were
then surveyed in March 1998 before benefits were received and then again in October 1998, June
48
1999 and November 1999. Families were queried on their family background, assets brought to
marriage, schooling, health status, parental attitudes, aspirations for their children, food and
nonfood consumption, and time allocation for all household members and self-employment
activities. Schools provided supplemental survey evidence on student achievement test scores.
In Mexico, completion of primary school is fairly comprehensive. The enrollment rate
for primary school is about 93 percent. However, rural children typically leave school after
completing the sixth grade, at which point the national enrollment rate drops to 55 percent. A
second decline in enrollment occurs at the tenth grade at which point only 58 percent of those
qualified to continue do so.
Benefits under the program are paid to the mother every two months. The size of the
stipend varies with the age of the child and the child’s gender, with higher stipends for girls. In
addition, financial aid for school supplies is paid twice each year.
Enrollment rates for treatment and control groups were then compared, controlling for
family and community factors. The impact of the program was found to be positive and
statistically significant. At the primary level, at which enrollment rates are already 90 to 94
percent, PROGRESA stipends increase attendance for boys by between 0.74 to 1.07 percent and
for girls by 0.96 to 1.45 percentage points.
The impact on secondary enrollment, however, is much more distinctive. Prior to the
beginning of the program, the enrollment rate in secondary school was 67 percent for girls and 73
percent for boys. Considering children in grades one through nine, the PROGRESA subsidy
increased the enrollment rate for girls by between 7.2 to 9.3 percentage points and by 3.5 to 5.8
percentage points for boys, as found by Schultz (2001). These preliminary results suggest that
PROGRESA will increase overall educational attainment for poor rural children in Mexico by
about ten percent, thereby raising adult income by eight percent.
3
Guerrero, Hidlago, Michocacan, Puebla, Queretero, San Luis Potosi, and Veracruz.
49
The impact of PROGRESA is largest for children making the transition to junior
secondary school. Enrollment rates for girls of this age receiving the education subsidy are 20
percent higher for girls and 10 percent higher for boys as compared to the control group.
As a byproduct of the empirical analysis, which examined other determinants of child
labor, the study also produced estimates of the effect of increasing the density of schools. In the
sample, 12 percent of children travel more than four kilometers to a junior secondary school. If
enough schools were built so that all children traveled less than four kilometers to their junior
secondary school, secondary-school enrollment for boys would rise by less than one-half of one
percent and enrollment for girls would rise by about one-third of one percent.
While PROGRESA has a substantial impact on most indicators of child welfare including
food consumption, physical stature, illness, school attendance and future income, the impact on
child labor is disappointing. There is only a modest decline in labor force participation rates for
children in the program, falling primarily on unpaid activities (Parker and Skoufias, 2000).
Neither do enrolled children in the program spend more time on schoolwork at home nor exhibit
higher achievement test scores than similar children who do not receive the stipend.
The efforts being made on behalf of children in the programs reviewed here are
impressive and encouraging, even if the results are sometimes mixed. For the most, however,
little has been done to compare the benefits from these programs to their costs. An exception is
Schultz (2001), who calculates the rate of return on the resources put into the PROGRESA
program. He finds a rate of return of 8% that is above and beyond both the role of the program in
reducing current poverty and any consumption benefits from education.
VI.
Labor Standards Initiatives in the International Arena
In the international arena, it is commonly argued that countries with poor labor practices
with regard to children should be sanctioned in some manner. Advocates are generally motivated
either by concern for the impact of low cost child labor on wages in industrialized countries or by
50
humanitarian concern for exploited children. We turn first to empirical evidence as to whether child
labor practices affect export performance or comparative advantage. We then turn to the impact that
trade policies are likely to have on the welfare of children.
Do National Labor Standards Alter Exports, Competitiveness or Comparative Advantage?
It is arguably the case that child labor may lower the wages of unskilled workers in
industrialized countries. A large volume of cheaply produced, unskilled-labor intensive exports
made possible with labor of children may have the effect of depressing the demand for such goods
produced in industrialized countries and, thereby, lower the wages of unskilled workers. To the
extent that child labor practices in developing countries have implications for industrialized country
workers, industrialized countries may seek reform or redress.
In order to determine whether child labor practices affect trade performance, several
researchers have examined a simple correlation between the existence and/or observance of corelabor standards and various measures of trade performance. For example, Mah (1997) analyzes the
trade performance of 45 developing countries and finds that each country’s export share of GDP is
strongly negatively correlated with rights to nondiscrimination, negatively correlated with freedomof-association rights and weakly negatively correlated with the right to organize and collective
bargaining.
However, such a correlation can have many reasons, and to gauge the marginal contribution
of core labor standards, one must compare each country's trade performance against a baseline
expectation as to what such a country should be trading given its factor endowments and other
determinants of trade. Rodrik (1996) provides an excellent example of how such analysis can be
undertaken.
He first considers the impact of core-labor standards on labor costs per worker in
manufacturing. He does this by calculating a regression using labor cost as the dependent variable
and per capita income and various measures of labor standards as the independent variables. In this
51
framework, per capita income is being used as a proxy for productivity in the economy. Labor
standards are measured in a variety of ways: total number of ILO conventions ratified; number of
ILO conventions ratified pertaining to labor standards; Freedom House indicators of civil liberties
and political rights; statutory hours worked; days of paid annual leave; the unionization rate; and an
indicator of child labor.
Rodrik finds that for the period 1985-1988, labor costs are overwhelmingly determined by
labor productivity. However, the number of ILO conventions ratified, Freedom House indicators of
democracy and the index of child labor are large and statistically significant, with laws regulating
child labor playing a particularly important role in statistically explaining labor costs.
Rodrik then turns to the determinants of comparative advantage in labor-intensive goods.
He uses the fraction of textiles and clothing exports in total exports as a proxy for measuring
comparative advantage in labor-intensive goods. As a theoretical matter, comparative advantage is
primarily determined by factor endowments. Therefore, the comparative advantage variable is
regressed on the independent variables of population-to-land ratio (a measure of the labor
endowment), average years of schooling in the population over 25 (a measure of the stock of human
capital) and the labor standards variables. The population and human capital variables have the
expected signs and are statistically significant. However, generally the labor standards variables,
while having the expected sign, are not statistically significant. The lone exception is statutory hours
worked. The longer the workweek, the stronger is the comparative advantage in textiles and
clothing.
Overall, the link from low labor standards in low-income countries to the wage of unskilled
workers in industrialized countries is not especially strong. Child labor practices in developing
countries are, at best, a secondary factor in determining comparative advantage and trade
performance.
52
Labor Protections and Humanitarian Concerns
While there may be some legitimate concern with the impact of labor practices on industrial
country workers, we may be equally motivated by humanitarian concerns for children. However,
while it is undoubtedly the case that voters in high-income countries are genuinely concerned with
the welfare of foreign children, it is not at all clear that these concerns can be constructively
addressed by applying trade disciplines. To understand the role that trade policy might play in
mediating humanitarian concerns with the process of production, it is important to distinguish
between two different forms in which these moral concerns might manifest themselves.
First, moral distaste may be a private good. For example, a consumer might prefer not to
consume goods produced by children or under poor working conditions. In this case, consumers
ought to have an opportunity to avoid goods produced in this manner, provided that they are willing
to pay the additional cost of production. In some cases, this might be accomplished by attaching a
product label detailing the conditions under which the good was produced (Freeman, 1994). But if
moral distaste is also a public good, consumers preferring that their fellow citizens also refrain from
such consumption, then one can make a case that countries that wish to do so should be allowed to
state a broad definition of immoral working conditions and, acting as a country, refuse to import such
goods.
However, this particular moral stance focuses only on alleviating the bad feeling that
consumers may have knowing they have consumed a good produced under unpleasant
circumstances. The welfare of the foreign workers themselves is not necessarily at issue. But if
consumers in high-income countries can exhaust their moral commitments simply by avoiding
consumption of goods produced in ways that they dislike contemplating, without regard for the
welfare of the children involved, then the humanitarian argument begins to lose some of its moral
gravity.4 If, by contrast, humanitarian and moral concerns focus on the welfare of the children
4
Product labeling does sometimes include provisions for improving the well-being of children who are
displaced from jobs as a result. For more on this see Brown et al. 2001.
53
themselves, rather than on the discomfort of the consumer, then the ability of trade sanctions to
address these concerns is highly limited.
In fact, trade sanctions in the face of weak child protections are as likely or even more likely
to harm children as they are to improve conditions. Maskus (1997) provides a detailed discussion of
this point. Consider the problem of child labor in the case of a small open economy in which the
export sector is adult labor-intensive, the import sector is capital-intensive, and a nontraded
intermediate input to the export sector is produced using child labor. The child's labor supply is
increasing in the child's wage and decreasing in the adult wage. The marginal child worker is the
youngest, since the opportunity cost in terms of foregone education falls as the child ages. In this
setting, a foreign tax imposed on goods produced by children can lead to the social optimum in the
sense of internalizing the external effect of child work on the well-being of western consumers.
Those children no longer working who receive an education may also be better off, although the fact
that they or their parents chose for them to work before makes this questionable. However, if, as a
consequence of the tax, the newly unemployed children live in a household with lower income, less
nutrition, and otherwise diminished life alternatives, the trade sanction has probably been counterproductive. Children who continue to work after the imposition of the tax are definitely worse off,
since the firms who employ children have to pay a tax. In a small open economy, a tax must lower
the after-tax wage of the working child.
One might wonder whether trade sanctions could be effective in the multiple-equilibrium
context of Basu and Van (1998), moving a country from a low-wage equilibrium to a high-wage
equilibrium that could then be sustained without the sanctions. Leaving aside the daunting empirical
question of how one could ever be sure that such multiple equilibria were present, Basu (1999) is
explicit in rejecting this as a basis for trade sanctions, arguing only that coordinated enforcement of
labor standards across countries might be appropriate. In fact, trade sanctions tend to reduce the
demand for labor in poor countries, not increase it, and if anything this would move a country to a
lower equilibrium, not a higher one.
54
The threat of sanctions will be particularly ineffective if the targeted country simply lacks
the resources to respond to the threat. For example, Rogers and Swinnerton (1999) estimate that if
GDP per worker falls below $5,020, families are so poor that they cannot survive without
contributions to family income from children. Thus, no matter how intense the demand for a
reduction in child labor, child labor practices will continue.
Furthermore, trade sanctions do little to address the underlying market failure that gives rise
to offending child labor practices. For example, as discussed above, capital market failure arguably
lies at the heart of the most egregious forms of child labor exploitation. If parents had access to
capital markets, they would school their children while transferring wealth from the future to the
present by borrowing against their own future income or the future income of their children.
However, lacking collateral and facing other capital market pathologies, the only device that parents
have available to them is to put their children to work. The end result, of course, is inadequate
human capital formation.
VII.
Conclusions
Concern for the welfare of working children has taken on a new importance in the
international arena over the last decade. While some participants in the global discussion focus
on the implications that working children might have on the rights and wages of workers in
industrialized countries, there is little evidence to support this concern. Although there are on the
order of 250 million children working worldwide, the value of their output is so small that it is
unlikely to have much of an effect on the international wage structure. Furthermore, most
children work in the informal sector or in home-work and, therefore, are not in direct competition
with unskilled workers in industrialized countries. Neither of these reasons means that child
labor has no effect at all on wages of other workers elsewhere, but those effects are surely small
compared to the effects on the children themselves. For this reason, much of the discussion with
regard to children focuses on the children, rather than on the implications for others.
55
When establishing policies with regard to children, it is essential therefore to have
reasonable confidence that policies put in place will actually improve the lives of the intended
beneficiaries. However, this is difficult to do given the wide array of factors that are affecting
parents and the work-school decisions they make for their children. For this reason, recent policy
initiatives have focused on providing incentives for families to choose education rather than
punishments. Attempts to use legal restrictions to affect household decisions take away options
that families are exercising and may leave children with worsened alternatives. By contrast,
incentive schemes open up new and improved alternatives to families without taking away
existing choices. Thus, if, in the presence of the incentive schemes, families choose less child
labor than in their absence, there is reason to believe that the policy has been effective in
improving the lot of children.
Although at this point the evidence is not clear that such incentive schemes will succeed
in significantly reducing the incidence of child labor, they still currently represent the best hope
for helping working children, while minimizing unintended negative effects. They deserve an
opportunity to succeed. Recent policy innovations are receiving earnest support from the World
Bank, the International Labour Organization, UNICEF and UNESCO, but they are in need of
further financial assistance, technical support and rigorous empirical evaluation.
The question then is which of the myriad policy configurations appears to have the
greatest potential to improve the lot of working children? First and foremost, both theory and
empirical evidence point very strongly to the role of capital-market failure in giving rise to
inefficient child labor. From a theoretical perspective, families without access to capital markets
may not be able to invest in their children even if it is optimizing for the family to do so.
As an empirical matter, it is generally the case that families with some household assets
such as older children, a mother with marketable skills, and assets associated with a household
enterprise are more likely to choose some education for their children than families without
assets. However, providing access to capital markets is a double-edge sword. While access to
56
capital markets may lead some families to borrow to finance an education for their children, there
are at least some cases in which households borrow to finance assets that are then combined with
more child labor. For this reason, those policies that offer assets in return for school attendance
provide the liquidity that make schooling possible while cutting off the option of taking children
out of school to work with household capital.
One of the striking results of the studies we have reviewed is that education subsidies that
might normally be expected to draw children out of the work force and into school do, in fact,
keep children in school. But we have very little statistically significant evidence that such
subsidies alter the parent’s decision as to whether the child should work. It is useful to consider
why this might be the case. Several possible explanations are suggested by the evidence.
First, the education subsidy may not be large enough to replace the child’s contribution to
family income. In this case, even if parents would like to put their children in school, they are too
poor to do so even given the subsidy. In the case of the Pakistani program, the subsidy was not
even replacing 20 percent or a working child’s earnings. This alone is enough to explain why
very few families returned full-time-working children to school. By contrast, the Brazilian
programs, PETI and Bolsa Escola, are means tested and provide subsidies that are typically quite
large, enough to pull a beneficiary family up to the poverty line.
Second, the families who do respond to the subsidy appear to be those with idle children,
neither working nor in school. The question, of course, is why are parents allowing their children
to remain idle? One possibility is that school quality (or the perception of it) is so poor that
parents see little point in going to the effort or expense of schooling their children. If school
quality is in fact poor, attending school may earn the subsidy but will have little impact on the
formation of the child’s human capital. Certainly, both empirical and anecdotal evidence point to
school quality as an important factor in a family’s work-school decision.
The issue of school quality may also help us understand the somewhat disappointing
impact that the PROGRESA program had on child labor in rural Mexico compared to the
57
stunning impact that PETI appears to have had on some working children in rural Brazil. Both
programs combine education subsidies with other forms of support for children. PROGRESA
emphasizes nutrition and health care. PETI emphasizes school structures and instructional
materials. It may be the case that improving school quality is more important than is health care
and nutrition for altering the family’s work-school calculation. However, it may also be the case
that the miraculous impact of PETI emerges in communities in which child labor is particularly
pervasive or the work that children are undertaking is extremely dangerous. Without the type of
careful empirical analysis applied to PROGRESA, it is difficult to draw conclusions.
Although the initial analysis of education subsidies appears to be disappointing, their
performance is not worse and, in some cases, far better than laws that mandate minimum years of
compulsory schooling and age of work. In fact, there appears to be very little evidence that such
regulations have more than a marginal impact on the age at which children leave school and begin
working. Therefore, a policy stance that requires the establishment and enforcement of child
labor standards across all countries is unlikely to be effective or improve the lot of children.
Neither is it reasonable to believe that trade sanctions leveled against countries with a
high rate of child labor are likely to make children better off. In fact, the threat of sanctions
against non-compliant countries is all too credible because of political forces within industrialized
countries that will promote them for a variety of reasons. But such threats are either disingenuous
or misguided, because sanctions are very likely to harm children rather than help them.
58
REFERENCES
Acemoglu, Daron and Joshua Angrist. 1999. “How Large are the Social Returns to Education?
Evidence from Compulsory Schooling Laws.” NBER Working Paper 7444.
Andvig, Jens Chr. 1997. “Child Labor in Sub-Saharan Africa – An Exploration,” NUPI Working
Paper no. 585, October.
Angrist, Joshua and Alan Krueger. 1991. “Does Compulsory Schooling Affect Schooling and
Earnings?” Quarterly Journal of Economics 106 (4): pp. 979-1014.
Anker, R. and S. Barge. 1998. Economics of Child Labour in Indian Industries. Geneva: ILO.
Anker, R and H. Melkas. 1996. “Economic Incentives for Children and Families to Eliminate or
Reduce Child Labour.” Geneva: ILO.
Baland, Jean-Marie and James A. Robinson. 2000. “Is Child Labor Inefficient?” Journal of
Political Economy, vol. 108, no. 4, p. 662-679.
Basu, Alaka Malwade. 1993. “Family Size and Child Welfare in an Urban Slum: Some
Disadvantages of Being Poor but Modern.” In Fertility, Family Size and Structure.
Cynthia B. Lloyd, ed. New York: Population Council.
Basu, Kaushik. 1999. “Child Labor: Cause, Consequence, and Cure, with Remarks on
International Labor Standards.” Journal of Economic Literature XXXVII (September):
pp. 1083-1119.
Basu, Kaushik and Pham Hoang Van. 1998. “The Economics of Child Labor.” American
Economic Review 88 (3): pp. 412-427.
Becker, Gary S. 1964. Human Capital. New York: Columbia University Press.
Becker, G.S. and H. G. Lewis. 1973. “On the Interaction between the Quantity and Quality of
Children.” Journal of Political Economy 81: S279-S288.
Bhalotra, S. and C. Heady. 1998. “Child Labor in Rural Pakistan and Ghana: Myths and Data.”
Working Paper. Department of Economics, University of Bristol, UK.
Bolin-Hort, Per. 1989. Work, Family, and the State: Child Labor and the Organization of
Production in the British Cotton Industry, 1780-1920. Lund: Lund U. Press.
Bonnet, M. 1993. “Child Labor in Africa.” International Labor Review 132, pp. 371-389.
Brown, Drusilla K., Alan V. Deardorff and Robert M. Stern. 2001. “U.S. Trade and Other Policy
Options and Programs to Deter Foreign Exploitation of Child Labor.” In: Topics in
Empirical International Economics, Magnus Blomstrom and Linda S. Goldberg (eds.).
Chicago and London: University of Chicago Press.
59
Brown, Martin, Jens Christiansen and Peter Philips. 1992. “The Decline of Child Labor in the US
Fruit and Vegetable Canning Industry: Law or Economics.” Business History Review, 66
(4) pp. 723-770.
Burra, Neera. 1995. Born to Work: Child Labor in India. Dehli: Oxford University Press.
Canagarajah, S. and H. Coulombe. 1998. “Child Labor and Schooling in Ghana.” Policy Research
Working Paper no. 1844. Washington, D.C.: The World Bank.
Canagarajah, Sudharshan and Helena Skyt Nielsen. 1999. “Child Labor and Schooling in Africa:
A Comparative Study, manuscript, World Bank.
Cartwright, Kimberly. 1999. “Child Labor in Colombia,” in The Policy Analysis of Child Labor,
A Comparative Study,” edited by Christiaan Grootaert and Harry Anthony Patrinos,
Washington, D.C.: World Bank.
Cartwright, Kimberly and Harry Anthony Patrinos. 1999. “ Child Labor in Urban Bolivia,” in The
Policy Analysis of Child Labor, A Comparative Study,” edited by Christiaan Grootaert
and Harry Anthony Patrinos, Washington, D.C.: World Bank.
Chernichovsky, D. 1985. “Socioeconomic and Demographic Aspects of School Enrollment and
Attendance in Rural Botswana. Economic Development and Cultural Change 32 (1), pp.
319-332.
Coulombe, H. 1998. “Child Labor and Education in Cote d’Ivoire.” Background paper. World
Bank: Washington D.C.
DeGraff, D.S., R.E. Bilsborrow, and A.N. Harriman. 1993. “The Implications of High Fertility
for Children’s Time Use in the Philippines.” In Fertility, Family Size, and Structure:
Consequences for Families and Children, C.B. Lloyd (ed.), Proceedings of a Population
Council Seminar. New York, June 9-10, 1992.
Dreze, Jean and Geeta Gandhi Kingdon. 2000. “School Participation in Rural India.” Manuscript,
Oxford University, April.
Duflo, Esther. 2000. “Schooling and Labor Market Consequences of School Construction in
Indonesia: Evidence from an Unusual Policy Experiment,” NBER Working Paper 7860.
Engerman, Stanly L. 2001. “The History and Political Economy of International Labor Standards,”
mimeo.
Freeman, Richard. 1994. “A Hard-Headed Look at Labour Standards,” in Werner Sengenberger and
Duncan Campbell, eds., International Labour Standards and Economic Interdependence.
Geneva, Switzerland: International Labour Office.
Garcia-Moreno, Mauricio. 1998. “Child Work and Education in Ecuador,” in Child Work and
Education, Five Case Studies from Latin America, edited by Maria Cristina Salazar and
Walter Alarcon Glasinovich, Florence: UNICEF.
60
Glasinovich, Walter Alarcon. 1998. ‘The Work and Education of Children in Peru,” in Child
Work and Education, Five Case Studies from Latin America, edited by Maria Cristina
Salazar and Walter Alarcon Glasinovich, Florence: UNICEF.
Glewwe, P. and H. Jacoby. 1994. “Student Achievement and Schooling Choice in Low-Income
Countries: Evidence from Ghana.” Journal of Human Resources 29, pp. 843-864.
Grootaert, Christiaan. 1999. “Child Labor in Cote d’Ivoire.” In The Policy Analysis of Child
Labor, A Comparative Study,” edited by Christiaan Grootaert and Harry Anthony
Patrinos, Washington, D.C.: World Bank.
Grootaert, Christiaan and Ravi Kanbur. 1995. “Child Labour: An Economic Perspective,”
International Labour Review 134(2), pp. 187-2003.
Gupta, Manash R. 1997. “Unemployment of Adult Labor and the Supply of Child Labor: A
Theoretical Analysis,” mimeo, Jadavpur U., Calcutta.
Hanushek, E.A. 1992. “The Trade-off between Child Quantity and Quality. Journal of Political
Economy 100 (1), pp. 647-660.
Hanushek, E. A. and V. Lavy. 1994. “School Quality, Achievement Bias, and Dropout Behavior
in Egypt.” LSMS Working Paper #107, World Bank: Washington, D.C.
Harmon, Colm and Ian Walker. 1995. “Estimates of the Economic Return to Schooling in the
U.K.” American Economic Review 85 (5), pp. 1278-1286.
IFPRI. 2000. “Is PROGRESA Working? Summary of the Results of an Evaluation by IFPRI.”
International Food Policy Research Institute, Washington, D.C.
ILO-IPEC. 1998a. Child Labour in Bangladesh: Its Context and Response to It. Dhaka:
ILO/IPEC, May.
ILO-IPEC. 1998b. Update Factory Monitoring per 01/05/90, ILO/IPEC, Memorandum to
Members of the Informal Steering Committee, Dhaka, May 13, 1998.
IBGE (1992), “Criancas e adolescentes: indicadores sociais:, Rio de Janeiro: IBGE, v. 4.
Jensen, P. and H.S. Nielsen. 1997. “Child Labour or School Attendance? Evidence from
Zambia.” Journal of Population Economics 10, pp. 407-424.
Krueger, Alan. 1996. "International Labor Standards and Trade,” in Michael Bruno and Boris
Pleskovic (eds.)Annual World Bank Conference on Development Economics, (Washington,
D.C.: The World Bank), pp. 281-302.
Lavinas, Lena and Maria Ligia Barbosa. 1999. “Combating Poverty by Increasing School
Attendance: The Case Study of Bolsa-Escola in Recife, manuscript.
Lavy, V. 1996. “School Supply Constraints and Children’s Educational Outcomes in Rural Ghana.”
Journal of Development Economics 51, pp. 291-314.
61
Levison, D. 1991. Children’s Labor Force Activity and Schooling in Brazil. Ph.D. Dissertation,
University of Michigan.
Levison, D., and R. Anker, S. Ashraf and S. Barge. 1996. Is Child Labor Really Necessary in India’s
Carpet Industry? Labor Market Paper No. 15. Geneva: ILO.
Levy, V. 1985. “Cropping Patterns, Mechanization, and Child Labor and Fertility Behaviour in a
Framing Economy: Rural Egypt.” Economic Development and Cultural Change 33: p777791.
Lloyd, C.B. 1994. “Investing in the Next Generation: The Implication of High Fertility at the Level
of the Family.” New York Population Council Working Paper No. 63.
Mah, J.S., 1997. “Core Labor Standards and Export Performance in Developing Countries.” World
Economy 20, no. 6 (September), pp. 773-785.
Maskus, Keith. 1997. "Should Core Labor Standards be Imposed Through International Trade
Policy?" World Bank Policy Research Working Paper No. 1817, August.
Moehling, Carolyn M. 1995. “The Intra-household Allocation of Resources and the Participation of
Children in Household Decision-Making: Evidence from Early Twentieth Century
America,” mimeo, Northwestern University.
Moehling, Carolyn M. 1999. “State Child Labor Laws and the Decline of Child Labor.” Explorations
in Economic History 36: pp. 72-106.
Montgomery, M., A. Kouame and R. Oliver. 1995. “The Tradeoff between Number of Children
and Child Schooling: Evidence from Cote d’Ivoire and Ghana.” LSMS Working Paper
112, Washington, D.C.: The World Bank.
Margo, Robert A. and T. Aldrich Finegan. 1996. “Compulsory Schooling Legislation and School
Attendance in Turn-of-the-Century America: A ‘Natural Experiment’ Approach.”
Economic Letters, 53, (October), : pp.103-110.
Ministerio de Educacion. 1993. “Censo Escolar”, Lima: Ministerio de Educacion (unpublished).
Nardinelli, Clark. 1990. Child Labor and the Industrial Revolution. Bloomington: Indiana U.
Press.
Nielsen, H.S. 1998. “Child Labor and School Attendance in Zambia: Two Joint Decisions.” CLS
WP 98-15. Centre for Labour Market and Social Research, Aarhus, Denmark.
OECD. 2000. International Trade and Core Labour Standards, Paris: OECD.
Parker, S. and E. Skoufias. 2000. “The Impact of PROGRESA on Work and Time Allocation.”
International Food Policy Research Institute, Washington, D.C., forthcoming.
Parsons, Donald O. and Claudia Goldin. 1989. “Parental Altruism and Self-Interest: Child Labor
Among Late Nineteenth-Century American Families.” Economic Inquiry 27(4): pp. 637659.
62
Patrinos, H. and G. Psacharopoulos. 1995. “Educational Performance and Child Labor in
Paraguay. International Journal of Educational Development 15 (1), pp. 47-60.
Patrinos, H. and G. Psacharopoulos. 1997. “Family Size, Schooling and Child Labor in Peru: An
Empirical Analysis.” Journal of Population Economics 10, pp. 387-405.
Powell, B. and L.C. Steelman. 1993. “The Educational Benefits of Being Spaced Out: Sibship
Density and Educational Progress. American Sociological Review 58, pp. 367-381.
Psacharopoulos, George. 1997. “Child Labor Versus Educational Attainment, Some Evidence
from Latin America.” Journal of Population Economics 10, pp. 377-386.
Psacharopoulos G. and A.M. Arriagada. 1989. “The Determinants of Early Age Human Capital
Formation: Evidence from Brazil. Economic Development and Cultural Change 37(4),
pp. 683-708.
Ranjan, Priya. 2001. “Credit Constraints and the Phenomenon of Child Labor.” Journal of
Development Economics 64, pp. 81-102.
Ravallion, Martin and Quentin Wodon. 1999. “Does Child Labor Displace Schooling? Evidence
on Behavioral Responses to an Enrollment Subsidy,” Washington, D.C.: The World
Bank.
Ray, Ranjan. 2000. “Analysis of Child Labour in Peru and Pakistan: A Comparative Study.”
Journal of Population Economics 13, pp. 3-19.
Rizzini, Irene, Irma Rizzini and Fernanda Rosa Borges. 1998. “Brazil: Children’s Strength is not
in Their Work,” in Child Work and Education, Five Case Studies from Latin America,
edited by Maria Cristina Salazar and Walter Alarcon Glasinovich, Florence: UNICEF.
Rodrik, Dani. 1996. "Labor Standards in International Trade: Do They Matter and What Do We Do
About Them," in Emerging Agenda For Global Trade: High States for Developing
Countries, edited by Robert Z. Lawrence, Dani Rodrik and John Whalley. Overseas
Development Council Essay No. 20, Washington, D.C.: Johns Hopkins University Press.
Rogers, Carol Ann and Kenneth A. Swinnerton. 1999. "Inequality, Productivity, and Child Labor:
Theory and Evidence," mimeo, Georgetown University.
Rogers, Carol Ann and Kenneth A. Swinnerton. 1999. “Inequality, Productivity, and Child Labor:
Theory and Evidence.” Mimeo, Georgetown University.
Rosenzweig, M.R. 1981. “Household and Non-Household Activities of Youth: Issues of
Modelling, Data and Estimation Strategies.” In: Child Work, Poverty and UnderDevelopment, G. Rodgers and G. Standing (eds.), Chapter 8, Geneva: ILO.
Rosenzweig, M.R. and K. Wolpin. 1980. “Testing the Quantit-Quality Fertility Model: The Use
of Twins as a Natural Experiment.” Econometrica 48, pp. 228-240.
Sakellariou, Chris and Ashish Lall. 1999. “Child Labor in the Philippines.” In The Policy
Analysis of Child Labor, A Comparative Study,” edited by Christiaan Grootaert and
Harry Anthony Patrinos, Washington, D.C.: World Bank.
63
San Martin, Orlando. 1996. “Child Labour and Socio-Economic Development.” In Bjorne
Grimsrud and Arne Melchior, (eds.) Child Labour and International Trade Policy, Paris:
OECD-DNMEs Workshop).
Scholliers, Peter. 1995. “Grown-ups, Boys and Girls in the Ghent Cotton Industry: The Voortman
Mills, 1835-1914.” Social History 20(2), pp. 201-218.
Schultz, Paul T. 2001. “School Subsidies for the Poor: Evaluating a Mexican Strategy for
Reducing Poverty.” International Food Policy Research Institute, FCND Discussion
Paper No. 102, March.
Spindel, Cheywa Rojza. 1985. O menor trabalhador: um assalariado registrado, Sao Paulo:
Nobel.
Strauss, J. and D. Thomas. 1995. “Human Resources: Empirical Modeling of Household and
Family Decisions,” in J. Behrman and T.N. Strinvasan (eds. Handbook of Development
Economics Volume 3, Amsterdam: North-Holland.
Turvay, Catalina and Elvia Maria Acuna. 1998. “Child Labour and Basic Education in
Colombia,” in Child Work and Education, Five Case Studies from Latin America, edited
by Maria Cristina Salazar and Walter Alarcon Glasinovich, Florence: UNICEF.
Tzannatos, Zafiris. 1996. “Child Labor and School Enrollment in Thailand in the 1990s.” Social
Protection, Human Development Network. The World Bank, Washington, D.C.
U.S. Department of Labor. 1998. By the Sweat and Toil of Children, Efforts to Eliminate Child
Labor, vol. 5, Washington D.C.
Wodon, Quentin. 1999. “Cost-Benefit Analysis of Food for Education in Bangladesh,” Policy
Research Working Paper, World Bank, Washington, D.C.
World Bank. 2001. Spectrum. Winter.
64
Table 1
Sequential Probit Analysis for Selected Countries in Latin America
Stage 1: Probability that a Child Works
Variables Significant at the 10% Level
Colombia
Bolivia
Peru
Rural
Urban
Year
Statistical Technique
Population Sampled
1993
1993
1993
Sequential Probit
7-17 years
12-17 years
7-17 years
Child Characteristics
Age
Male
7.72
9.03
7.16
10.86
Parent Characteristics
Age Head of HH
Father’s Education
Mother’s Education
Mother Working?
1.98
-1.98
-1.95
7.30
-1.02
-1.79
3.84
-0.37
-10.81
-18.68
-4.63
HH Characteristics
Log Expenditures
Rural
Indigenous
Family Enterprise
1991
Logit
7-18 years
1.76
3.62
-0.40
15.08
7.76
7.77
Siblings
Aged 0-6
Sisters 10-15
16+
Brothers 6-9
10-15
16+
-13.53
-4.88
-6.69
-16.72
-5.15
-14.58
-8.91
-4.76
-0.29
-17.26
-8.77
-3.96
Cost of Schooling
-8.41
-7.59
6.16
Observations
1829
9821
4730
2.70
Entries are probability derivatives at the mean of the explanatory variables.
Sources: Adapted from Cartwright (1999), Cartwright and Patrinos (1999) and
Patrinos and Psacharopoulos (1997).
1727
65
Table 2
Sequential Probit and Logit Analysis for Selected Countries in Latin America
Stage 2: Probability of Full-Time Work
Variables Significant at the 10% Level
Colombia
Bolivia
Peru
Rural
Urban
Year
Statistical Technique
Dependent Variable
Population Sampled
Child Characteristics
Age
Male
Parent Characteristics
Father’s Education
Mother’s Education
Mother Working?
Father Union Member
HH Characteristics
Log Expenditures
Family Enterprise
No. of Rooms
Rural
Indigenous language
Siblings
Sisters 0-5 years
10-15
16+
Brothers 0-5 years
10-15
16+
Cost of Schooling
Private School
Observations
1993
Probability of
Fulltime Work
7-17 years
6.78
1993
Sequential Probit
Probability of
Fulltime Work
12-17 years
4.59
1993
Probability of
Fulltime Work
7-17 years
1991
Logit
Age-Grade
Distortion
7-18 years
3.64
-31.35
-13.75
-12.95
-1.10
-1.21
-8.25
-8.56
-14.11
-1.53
-3.10
-33.48
-30.47
-34.88
-27.42
-2.84
11.11
-9.05
-12.87
-9.54
6.15*
2.55*
2.84*
-10.71
12.89
8.51
624
22.20
2.67
-5.70
-7.06
-3.90
-1.02
-21.49
8.92
1915
590
*Brothers and sisters.
Entries are probability derivatives at the mean of the explanatory variables.
Sources: Adapted from Cartwright (1999), Cartwright and Patrinos (1999) and Patrinos and
Psacharopoulos (1997).
-6.93
1727
66
Table 3
Probability of Child Work and Schooling for Selected Countries in Africa
Variables Significant at the 10% Level
Ghana
Cote d’Ivoire
Year
Dependant Variable
Sample Population
Statistical Method
Child Characteristics
Age
Female
Grade
Married
Parent Characteristics
HH Head Age
Father’s Education
Mother’s Education
Mother Working?
Girl with Working Mother
Father in Formal Sector
Proportion of Non-Head
Working
Father in HH
Head Not Working
HH Characteristics
Log Expenditures/capita
Assets
Savings
Land
Family Enterprise
Protestant
Catholic
Other Christian
Muslim
Agriculture/Animals
Rural
Rain Forest
Poor
Other HH Members
Boys 6-9
10-15
16-17
1991-1992
Probability of
Some
Some
Work School
School
Only
7-14 year olds
Bivariate Probit
BetaCoefficient
Urban 7-17 year olds
Sequential Probit and Bivariate Logit
Probability Derivatives at the
Variable Mean
0.85
0.23
0.66
-0.37
-0.29
1.26
0.75
-29.80
2.91
1.48
1993
Probability of
School and
Work
Work Only
21.05
-16.40
9.33
-15.58
2.73
1.84
3.54
Zambia
1993
Probability
of School
7-14 year
olds
BetaCoefficient
-0.20
-0.22
0.30
-2.94
0.06
10.63
18.25
0.15
-1.59
0.32
-0.29
0.28
-0.22
0.26
0.45
-10.04
1.55
0.16
0.06
-0.36
-12.62
0.61
0.46
0.40
0.17
-0.34
-0.26
0.62
9.22
29.22
-8.63
-12.51
4.6
3.65
13.93
0.55
0.30
67
Girls 6-9
10-15
15+
Aged 60+
Cost of Schooling
Distance to School
Households
6.33
0.12
4.39
9.02
-0.21
0.16
-0.11
0.11
-0.03
3811
1177
Sources: Canagarajah and Coloumbe (1998), Grootaert (1999) and Nielsen (1998)
6372
68
Table 4
Sequential Probit Analysis for Selected Countries in Asia
Variables Significant at the 10% Level
India
Philippines
Boys
Girls
Year
Statistical Technique
Age of Sample Population
Dependent Variable
Child Characteristics
Age
5 years old
8 years old
10 years old
11 years old
12 years old
Parent Characteristics
Age Head of HH
Father’s Education
Mother’s Education
Mother Working?
Mother of Girl Working?
HH Head of Girl Working?
Female HH Head Working?
HH Characteristics
Poor
Rural
Member of Caste or Tribe
Other Backward Caste
Believe Girls Should be Educated
Male Head of HH
Married Head of HH
Assets
Family Enterprise
Cow/Goat
Siblings
#children/#adults in HH
Binary Logit
5-12 years old
Probability of
Enrollment
Sequential Probit
10-17 years old
Probability of
Work
0.024
-0.233
-0.317
-0.101
-0.117
-0.131
-0.175
0.009
0.020
0.025
0.011
0.019
-0.020
0.029
0.038
0.024
-0.044
-0.042
0.097
-0.119
-0.116
0.303
0.031
-0.027
0.006
0.278
-0.011
-0.050
School Characteristics
School Lunch Program
Building Waterproof
School Open Previous Week
0.149
0.146
0.111
Village Characteristics
Elect., P.O., Water, Phone
0.036
69
Women’s Association
Observations
0.102
1405
1067
23062
Entries are beta-coefficients of the explanatory variables.
Sources: Adapted from Dreze and Kingdon (2000) and Sakalariou and Lall
(1999).
70
TABLE 5
CHILD LABOR AND EDUCATION
LABOR FORCE PARTICIPATION RATES, MINMUM AGE OF WORK
AND COMPULSORY EDUCATION
Region
Child Labor Force
Participation
Age Range
Rate
Minimum Age for Work
Basic
Hazardous
Compulsory
Education
Ages
Africa
Egypt
Kenya
South Africa
Tanzania
5-14
6-14
10-14
10-14
10-14
41.0
12.0
41.3
4.3
39.5
14
16
15
12-18
15-17
16-18
18
18
Asia
Bangladesh
India
Nepal
Pakistan
Philippines
Thailand
5-14
5-14
5-14
5-14
5-14
5-14
10-14
21.0
19.1
5.4
41.7
8.0
10.6
16.2
12-15
14
14
14
15
15
18
14-18
16
14-21
18
18
6-11
6-11
Latin America
Brazil
Guatemala
Mexico
Nicaragua
Peru
5-14
5-14
7-14
12-14
10-14
6-14
17.0
12.8
4.1
17.3
9.9
4.1
14
14
14
14
12-16
18-21
16
16-18
18
18
7-14
6-15
6-14
7-12
6-16
Europe
Turkey
5-14
6-14
12.6
15
18
6-13
Source: Adapted from U.S. DOL (1998).
6-13
7-15
7-13
6-10
71
TABLE 6
PERCENT OF CHILDREN LEAVING SCHOOL BY AGE FOR SELECTED COUNTRIES
1959-74 BIRTH COHORT
School
Mexico Argentina Brazil
Leaving Age
12 or younger
25.2
6.6
80.2
13
1.3
5.9
5.8
14
1.8
5.9
3.7
15
5.1
3.5
3.5
16
4.0
4.5
2.1
17
3.8
10.8
2.1
18
7.8
16.0
1.0
19
5.3
8.0
0.7
20
6.5
4.2
0.4
21 or older
39.1
34.5
0.4
Adapted from Krueger (1997).
Nigeria
4.2
2.0
7.5
8.6
6.4
6.6
14.3
7.8
8.5
34.1
Chile
2.4
2.0
2.7
3.7
5.2
12.3
21.9
10.4
8.1
31.3
India
40.5
4.2
5.1
3.2
3.6
4.3
6.5
4.9
4.5
23.3
Portugal
17.4
7.4
10.7
4.6
5.4
4.6
7.8
0.0
10.6
31.6
Britain
0.0
0.4
0.0
5.3
50.1
10.3
11.2
0.0
7.4
15.2
USA
0.6
0.2
0.0
1.8
2.6
15.0
29.7
4.4
4.9
40.8
WORKING PAPER SERIES 2001
ase.tufts.edu/econ/papers
2001-01
EGGLESTON, Karen. Multitasking, Competition and Provider
Payment.
2001-02
IOANNIDES, Yannis M. Interactive Property Valuations.
2001-03
IOANNIDES, Yannis M. Neighborhood Income Distributions.
2001-04
IOANNIDES, Yannis M. Topologies of Social Interactions.
2001-05
KUTSOATI, Edward and Jan ZABOJNIK. Durable Goods
Monopoly, Learning-by-doing and “Sleeping Patents”.
2001-06
FULLERTON, Don and Gilbert E. METCALF. Tax Incidence.
2001-07
CHADHA, Rajesh, Drusilla K. BROWN, Alan V. DEARDORFF,
and Robert M. STERN. Computational Analysis of the Impact on
India of the Uruguay Round and the Forthcoming WTO Trade
Negotiations.
2001-08
BROWN, Drusilla K., Alan V. Deardorff, and Robert M. STERN.
CGE Modeling and Analysis of Multilateral and Regional
Negotiating Options.
2001-09
BROWN, Drusilla K. Child Labor in Latin America: Policy and
Evidence.
2001-10
BROWN, Drusilla K., Alan V. DEARDORFF, and Robert M.
STERN. Child Labor: Theory, Evidence and Policy.
2001-11
BROWN, Drusilla K., Alan V. DEARDORFF, and Robert M.
STERN. Impacts on NAFTA Members of Multilateral and
Regional Trading Arrangements and Initiatives and Harmonization
of NAFTA’s External Tariffs.
2001-12
BROWN, Drusilla K., Alan V. DEARDORFF, and Robert M.
STERN. Multilateral, Regional, and Bilateral Trade-Policy
Options for the United States and Japan.
2001-13
BROWN, Drusilla K. Labor Standards: Where Do They Belong on
the International Trade Agenda?